Buckets:
| <meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Action Representations","local":"action-representations","sections":[{"title":"Joint Space vs End-Effector Space","local":"joint-space-vs-end-effector-space","sections":[{"title":"Joint Space","local":"joint-space","sections":[],"depth":3},{"title":"End-Effector (EE) Space","local":"end-effector-ee-space","sections":[],"depth":3},{"title":"Converting Between Spaces","local":"converting-between-spaces","sections":[],"depth":3}],"depth":2},{"title":"Absolute, Relative, and Delta Actions","local":"absolute-relative-and-delta-actions","sections":[{"title":"Absolute Actions (LeRobot default)","local":"absolute-actions-lerobot-default","sections":[],"depth":3},{"title":"Relative Actions (used by OpenPI / pi0)","local":"relative-actions-used-by-openpi--pi0","sections":[],"depth":3},{"title":"Delta Actions (sequential differences)","local":"delta-actions-sequential-differences","sections":[],"depth":3},{"title":"Visual Comparison","local":"visual-comparison","sections":[],"depth":3}],"depth":2},{"title":"Using Relative Actions in LeRobot","local":"using-relative-actions-in-lerobot","sections":[{"title":"How it works","local":"how-it-works","sections":[],"depth":3},{"title":"Enabling relative actions for the pi family (pi0, pi0.5, pi0_fast)","local":"enabling-relative-actions-for-the-pi-family-pi0-pi05-pi0fast","sections":[],"depth":3},{"title":"Combining relative actions with RTC","local":"combining-relative-actions-with-rtc","sections":[],"depth":3},{"title":"Combining relative actions with EE space","local":"combining-relative-actions-with-ee-space","sections":[],"depth":3}],"depth":2},{"title":"Processing Pipeline Summary","local":"processing-pipeline-summary","sections":[],"depth":2},{"title":"References","local":"references","sections":[],"depth":2}],"depth":1}"> | |
| <link href="/docs/lerobot/pr_3313/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/entry/start.d3f1c0f3.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/scheduler.eb244325.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/singletons.1f33814c.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/index.3c23fb4b.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/paths.17f05d75.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/entry/app.04bb7687.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/preload-helper.b00aacbc.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/index.3fe63ad3.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/nodes/0.07fbe93e.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/each.e59479a4.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/nodes/3.137ba94a.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/CopyLLMTxtMenu.d0c64540.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/MermaidChart.svelte_svelte_type_style_lang.6453902c.js"> | |
| <link rel="modulepreload" href="/docs/lerobot/pr_3313/en/_app/immutable/chunks/CodeBlock.48dd2cc2.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"Action Representations","local":"action-representations","sections":[{"title":"Joint Space vs End-Effector Space","local":"joint-space-vs-end-effector-space","sections":[{"title":"Joint Space","local":"joint-space","sections":[],"depth":3},{"title":"End-Effector (EE) Space","local":"end-effector-ee-space","sections":[],"depth":3},{"title":"Converting Between Spaces","local":"converting-between-spaces","sections":[],"depth":3}],"depth":2},{"title":"Absolute, Relative, and Delta Actions","local":"absolute-relative-and-delta-actions","sections":[{"title":"Absolute Actions (LeRobot default)","local":"absolute-actions-lerobot-default","sections":[],"depth":3},{"title":"Relative Actions (used by OpenPI / pi0)","local":"relative-actions-used-by-openpi--pi0","sections":[],"depth":3},{"title":"Delta Actions (sequential differences)","local":"delta-actions-sequential-differences","sections":[],"depth":3},{"title":"Visual Comparison","local":"visual-comparison","sections":[],"depth":3}],"depth":2},{"title":"Using Relative Actions in LeRobot","local":"using-relative-actions-in-lerobot","sections":[{"title":"How it works","local":"how-it-works","sections":[],"depth":3},{"title":"Enabling relative actions for the pi family (pi0, pi0.5, pi0_fast)","local":"enabling-relative-actions-for-the-pi-family-pi0-pi05-pi0fast","sections":[],"depth":3},{"title":"Combining relative actions with RTC","local":"combining-relative-actions-with-rtc","sections":[],"depth":3},{"title":"Combining relative actions with EE space","local":"combining-relative-actions-with-ee-space","sections":[],"depth":3}],"depth":2},{"title":"Processing Pipeline Summary","local":"processing-pipeline-summary","sections":[],"depth":2},{"title":"References","local":"references","sections":[],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <div class="items-center shrink-0 min-w-[100px] max-sm:min-w-[50px] justify-end ml-auto flex" style="float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"><div class="inline-flex rounded-md max-sm:rounded-sm"><button class="inline-flex items-center gap-1 h-7 max-sm:h-7 px-2 max-sm:px-1.5 text-sm font-medium text-gray-800 border border-r-0 rounded-l-md max-sm:rounded-l-sm border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-live="polite"><span class="inline-flex items-center justify-center rounded-md p-0.5 max-sm:p-0 hover:text-gray-800 dark:hover:text-gray-200"><svg class="sm:size-3.5 size-3" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg></span> <span>Copy page</span></button> <button class="inline-flex items-center justify-center w-6 max-sm:w-5 h-7 max-sm:h-7 disabled:pointer-events-none text-sm text-gray-500 hover:text-gray-700 dark:hover:text-white rounded-r-md max-sm:rounded-r-sm border border-l transition border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-haspopup="menu" aria-expanded="false" aria-label="Open copy menu"><svg class="transition-transform text-gray-400 overflow-visible sm:size-3.5 size-3 rotate-0" width="1em" height="1em" viewBox="0 0 12 7" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M1 1L6 6L11 1" stroke="currentColor"></path></svg></button></div> </div> <h1 class="relative group"><a id="action-representations" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#action-representations"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Action Representations</span></h1> <p data-svelte-h="svelte-19ydbm0">This guide explains the different ways robot actions can be represented in LeRobot, how they relate to each other, and when to use each one.</p> <h2 class="relative group"><a id="joint-space-vs-end-effector-space" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#joint-space-vs-end-effector-space"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Joint Space vs End-Effector Space</span></h2> <p data-svelte-h="svelte-11qbh51">Before discussing action representations, it helps to understand the two coordinate spaces actions can live in.</p> <h3 class="relative group"><a id="joint-space" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#joint-space"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Joint Space</span></h3> <p data-svelte-h="svelte-1e2inq4">Joint-space actions directly specify target positions for each motor. For a 6-DOF arm with a gripper, a joint-space action might look like:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-string">action</span> <span class="hljs-string">=</span> [<span class="hljs-attr">shoulder_pan:</span> <span class="hljs-number">45.0</span>, <span class="hljs-attr">shoulder_lift:</span> <span class="hljs-number">-20.0</span>, <span class="hljs-attr">elbow:</span> <span class="hljs-number">-30.0</span>, <span class="hljs-attr">wrist_pitch:</span> <span class="hljs-number">10.0</span>, <span class="hljs-attr">wrist_roll:</span> <span class="hljs-number">0.0</span>, <span class="hljs-attr">wrist_yaw:</span> <span class="hljs-number">5.0</span>, <span class="hljs-attr">gripper:</span> <span class="hljs-number">0.8</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-4n2rwf">Joint space is the default in LeRobot. It is simple, requires no kinematics model, and maps directly to motor commands. Most beginner setups (SO-100, Koch) use joint-space actions.</p> <h3 class="relative group"><a id="end-effector-ee-space" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#end-effector-ee-space"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>End-Effector (EE) Space</span></h3> <p data-svelte-h="svelte-1y9ifw5">End-effector-space actions specify the desired position and orientation of the robot’s tool tip (gripper) in Cartesian coordinates:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-string">action</span> <span class="hljs-string">=</span> [<span class="hljs-attr">x:</span> <span class="hljs-number">0.25</span>, <span class="hljs-attr">y:</span> <span class="hljs-number">-0.10</span>, <span class="hljs-attr">z:</span> <span class="hljs-number">0.15</span>, <span class="hljs-attr">wx:</span> <span class="hljs-number">0.0</span>, <span class="hljs-attr">wy:</span> <span class="hljs-number">0.0</span>, <span class="hljs-attr">wz:</span> <span class="hljs-number">0.1</span>, <span class="hljs-attr">gripper:</span> <span class="hljs-number">0.8</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-241nfv">EE space is more intuitive for tasks like pick-and-place because it directly describes where the gripper should go, but it requires a kinematics model (URDF) to convert between EE poses and joint angles.</p> <h3 class="relative group"><a id="converting-between-spaces" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#converting-between-spaces"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Converting Between Spaces</span></h3> <p data-svelte-h="svelte-16ezse">LeRobot provides processor steps for converting between joint and EE spaces using forward and inverse kinematics. These are built on top of <code>RobotKinematics</code>, which loads a URDF model of your robot.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> lerobot.model.kinematics <span class="hljs-keyword">import</span> RobotKinematics | |
| <span class="hljs-keyword">from</span> lerobot.robots.so_follower.robot_kinematic_processor <span class="hljs-keyword">import</span> ( | |
| ForwardKinematicsJointsToEE, | |
| InverseKinematicsEEToJoints, | |
| ) | |
| kinematics = RobotKinematics( | |
| urdf_path=<span class="hljs-string">"./SO101/so101_new_calib.urdf"</span>, | |
| target_frame_name=<span class="hljs-string">"gripper_frame_link"</span>, | |
| joint_names=[<span class="hljs-string">"shoulder"</span>, <span class="hljs-string">"elbow"</span>, <span class="hljs-string">"wrist_pitch"</span>, <span class="hljs-string">"wrist_roll"</span>, <span class="hljs-string">"wrist_yaw"</span>], | |
| ) | |
| <span class="hljs-comment"># Joints → EE (for observations: "where is my gripper?")</span> | |
| fk_step = ForwardKinematicsJointsToEE(kinematics=kinematics, motor_names=[...]) | |
| <span class="hljs-comment"># EE → Joints (for actions: "move my gripper here")</span> | |
| ik_step = InverseKinematicsEEToJoints(kinematics=kinematics, motor_names=[...])<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-119gvsg">See <a href="https://github.com/huggingface/lerobot/tree/main/examples/so100_to_so100_EE" rel="nofollow"><code>examples/so100_to_so100_EE/</code></a> for a complete working example of recording, replaying, and evaluating with EE-space actions on an SO-100 arm.</p> <h2 class="relative group"><a id="absolute-relative-and-delta-actions" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#absolute-relative-and-delta-actions"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Absolute, Relative, and Delta Actions</span></h2> <p data-svelte-h="svelte-z7skmg">Regardless of whether you work in joint space or EE space, the action values can be expressed in three different ways. The terminology follows <a href="https://arxiv.org/abs/2402.10329" rel="nofollow">UMI (Chi et al., 2024)</a>.</p> <h3 class="relative group"><a id="absolute-actions-lerobot-default" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#absolute-actions-lerobot-default"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Absolute Actions (LeRobot default)</span></h3> <p data-svelte-h="svelte-7uwqu1">Each action specifies the target position directly.</p> <p data-svelte-h="svelte-9ik9fg"><strong>Example</strong> (joint space, chunk of 4):</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->current_state = [<span class="hljs-number">45.0</span>, <span class="hljs-number">-30.0</span>, <span class="hljs-number">10.0</span>] | |
| action_chunk = [ | |
| [<span class="hljs-meta">46.0, -29.0, 11.0</span>], <span class="hljs-meta"># go to 46, -29, 11</span> | |
| [<span class="hljs-meta">47.5, -27.0, 12.0</span>], <span class="hljs-meta"># go to 47.5, -27, 12</span> | |
| [<span class="hljs-meta">49.0, -25.0, 13.5</span>], <span class="hljs-meta"># go to 49, -25, 13.5</span> | |
| [<span class="hljs-meta">50.0, -24.0, 15.0</span>], <span class="hljs-meta"># go to 50, -24, 15</span> | |
| ]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-7r6n36">Each value is a target position in the robot’s coordinate frame. Simple and direct, but requires a consistent global coordinate frame. This is the default in LeRobot.</p> <h3 class="relative group"><a id="relative-actions-used-by-openpi--pi0" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#relative-actions-used-by-openpi--pi0"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Relative Actions (used by OpenPI / pi0)</span></h3> <p data-svelte-h="svelte-1elzjuw">Each action in the chunk is an offset from the <strong>current state at the moment of prediction</strong>. All actions in the chunk share the same reference point:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->current_state = [<span class="hljs-number">45.0</span>, <span class="hljs-number">-30.0</span>, <span class="hljs-number">10.0</span>] | |
| relative_chunk = [ | |
| <span class="hljs-string"> [1.0, 1.0, 1.0]</span>, # +<span class="hljs-number">1</span> <span class="hljs-keyword">from</span> current → target <span class="hljs-number">46</span>, <span class="hljs-number">-29</span>, <span class="hljs-number">11</span> | |
| <span class="hljs-string"> [2.5, 3.0, 2.0]</span>, # +<span class="hljs-number">2.5</span> <span class="hljs-keyword">from</span> current → target <span class="hljs-number">47.5</span>, <span class="hljs-number">-27</span>, <span class="hljs-number">12</span> | |
| <span class="hljs-string"> [4.0, 5.0, 3.5]</span>, # +<span class="hljs-number">4</span> <span class="hljs-keyword">from</span> current → target <span class="hljs-number">49</span>, <span class="hljs-number">-25</span>, <span class="hljs-number">13.5</span> | |
| <span class="hljs-string"> [5.0, 6.0, 5.0]</span>, # +<span class="hljs-number">5</span> <span class="hljs-keyword">from</span> current → target <span class="hljs-number">50</span>, <span class="hljs-number">-24</span>, <span class="hljs-number">15</span> | |
| ]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-fzxfo8">The conversion is straightforward: <code>relative = absolute - current_state</code>. To recover absolute: <code>absolute = relative + current_state</code>.</p> <p data-svelte-h="svelte-2n2ond"><strong>Why use relative actions?</strong> The model learns to predict offsets centered around zero, which is easier to normalize and leads to more stable training. Because every chunk references the same current state, there is no error accumulation across chunks.</p> <h3 class="relative group"><a id="delta-actions-sequential-differences" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#delta-actions-sequential-differences"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Delta Actions (sequential differences)</span></h3> <p data-svelte-h="svelte-1q782mp">Each action is an offset from the <strong>previous action</strong> (or from the current state for the first step):</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->current_state = [<span class="hljs-number">45.0</span>, <span class="hljs-number">-30.0</span>, <span class="hljs-number">10.0</span>] | |
| delta_chunk = [ | |
| [<span class="hljs-meta">1.0, 1.0, 1.0</span>], <span class="hljs-meta"># current → 46, -29, 11</span> | |
| [<span class="hljs-meta">1.5, 2.0, 1.0</span>], <span class="hljs-meta"># previous action → 47.5, -27, 12</span> | |
| [<span class="hljs-meta">1.5, 2.0, 1.5</span>], <span class="hljs-meta"># previous action → 49, -25, 13.5</span> | |
| [<span class="hljs-meta">1.0, 1.0, 1.5</span>], <span class="hljs-meta"># previous action → 50, -24, 15</span> | |
| ]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1odk8o8">Here each step is relative to the one before it. To recover absolute positions you must sum all previous deltas, which means errors accumulate over time. UMI explicitly argues against this representation for this reason.</p> <h3 class="relative group"><a id="visual-comparison" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#visual-comparison"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Visual Comparison</span></h3> <p data-svelte-h="svelte-pbqouw">The figure below (based on a figure from <a href="https://arxiv.org/abs/2402.10329" rel="nofollow">UMI, Chi et al., 2024</a>) illustrates the key difference. With <strong>relative trajectory</strong>, every action in the chunk points back to the same origin (current state), so a new inference step cleanly resets the reference. With <strong>delta</strong>, each action depends on the previous one, so errors accumulate. <strong>Absolute</strong> actions require a consistent global coordinate frame.</p> <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/lerobot/action_representations_umi.png" alt="Relative Trajectory as Action Representation (UMI, Chi et al., 2024)" width="85%"> <h2 class="relative group"><a id="using-relative-actions-in-lerobot" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#using-relative-actions-in-lerobot"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Using Relative Actions in LeRobot</span></h2> <p data-svelte-h="svelte-vdluzd">LeRobot provides <code>RelativeActionsProcessorStep</code> to convert between absolute and relative actions inside the processor pipeline. This is how pi0, pi0.5, and pi0_fast support relative actions.</p> <blockquote data-svelte-h="svelte-1xe88x3"><p><strong>Note:</strong> All pi models (pi0, pi0.5, pi0<em>fast) apply relative conversion _before</em> normalization (<code>relative → normalize</code>), so the normalizer always sees delta (relative) values. This means <strong>relative action stats are required</strong> for all of them when training with <code>use_relative_actions=true</code>. In pi0_fast the <code>RelativeActionsProcessorStep</code> only modifies the action — the state observation is unchanged — so <code>NormalizerProcessorStep</code> still runs before the state tokenizer and the tokenizer continues to receive normalized state as expected.</p></blockquote> <h3 class="relative group"><a id="how-it-works" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#how-it-works"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>How it works</span></h3> <p data-svelte-h="svelte-1ij4x7z">During <strong>training</strong> (preprocessing), actions are converted from absolute to relative before the model sees them:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->raw <span class="hljs-keyword">absolute</span> action → RelativeActionsProcessorStep → normalize → model<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-a0sv07">During <strong>inference</strong> (postprocessing), model predictions are converted back to absolute before being sent to the robot:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">model</span> output → unnormalize → AbsoluteActionsProcessorStep → robot<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-b2xqrn">The <code>AbsoluteActionsProcessorStep</code> reads the cached current state from its paired <code>RelativeActionsProcessorStep</code>, so the two must be wired together (handled automatically by the policy factory).</p> <h3 class="relative group"><a id="enabling-relative-actions-for-the-pi-family-pi0-pi05-pi0fast" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#enabling-relative-actions-for-the-pi-family-pi0-pi05-pi0fast"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Enabling relative actions for the pi family (pi0, pi0.5, pi0_fast)</span></h3> <p data-svelte-h="svelte-5x5i9h"><strong>Step 1</strong>: Precompute relative action statistics for your dataset:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->lerobot-edit-dataset \ | |
| --repo_id your_dataset \ | |
| --operation.type recompute_stats \ | |
| --operation.relative_action <span class="hljs-literal">true</span> \ | |
| --operation.chunk_size 50 \ | |
| --operation.relative_exclude_joints <span class="hljs-string">"['gripper']"</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-130ix7t"><strong>Step 2</strong>: Train with relative actions enabled:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->lerobot-train \ | |
| --dataset.repo_id=your_dataset \ | |
| --policy.type=pi0 \ | |
| --policy.use_relative_actions=<span class="hljs-literal">true</span> \ | |
| --policy.relative_exclude_joints=<span class="hljs-string">'["gripper"]'</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1tl3o3d">The <code>relative_exclude_joints</code> parameter specifies joints that should remain in absolute space. For example, gripper commands are typically binary (open/close) and don’t benefit from relative encoding.</p> <h3 class="relative group"><a id="combining-relative-actions-with-rtc" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#combining-relative-actions-with-rtc"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Combining relative actions with RTC</span></h3> <p data-svelte-h="svelte-1teo0n6"><a href="https://arxiv.org/abs/2506.07339" rel="nofollow">RTC</a> runs policy inference at high frequency and sends actions to the robot as they are predicted rather than waiting for a full chunk. Relative actions and RTC are fully compatible: because every chunk in relative mode references the <strong>same</strong> current state (captured at the start of inference), each predicted action in the chunk remains a valid offset even if the robot has already moved. No special handling is needed — <code>RelativeActionsProcessorStep</code> caches the state once per inference call and <code>AbsoluteActionsProcessorStep</code> applies it to every action in the streamed output.</p> <h3 class="relative group"><a id="combining-relative-actions-with-ee-space" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#combining-relative-actions-with-ee-space"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Combining relative actions with EE space</span></h3> <p data-svelte-h="svelte-14jteoq">Relative actions work in both joint space and EE space. For example, if your dataset stores EE actions, relative encoding converts them to offsets from the current EE pose:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-attr">current_ee_state</span> = [x: <span class="hljs-number">0.25</span>, y: -<span class="hljs-number">0.10</span>, z: <span class="hljs-number">0.15</span>, gripper: <span class="hljs-number">0.8</span>] | |
| <span class="hljs-attr">absolute_ee_chunk</span> = [ | |
| [<span class="hljs-number">0.26</span>, -<span class="hljs-number">0.09</span>, <span class="hljs-number">0.16</span>, <span class="hljs-number">0.8</span>], | |
| [<span class="hljs-number">0.28</span>, -<span class="hljs-number">0.07</span>, <span class="hljs-number">0.18</span>, <span class="hljs-number">0.8</span>], | |
| ] | |
| <span class="hljs-attr">relative_ee_chunk</span> = [ | |
| [<span class="hljs-number">0.01</span>, <span class="hljs-number">0.01</span>, <span class="hljs-number">0.01</span>, <span class="hljs-number">0.0</span>], <span class="hljs-comment"># offset from current EE pose</span> | |
| [<span class="hljs-number">0.03</span>, <span class="hljs-number">0.03</span>, <span class="hljs-number">0.03</span>, <span class="hljs-number">0.0</span>], <span class="hljs-comment"># offset from current EE pose</span> | |
| ]<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="processing-pipeline-summary" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#processing-pipeline-summary"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Processing Pipeline Summary</span></h2> <p data-svelte-h="svelte-1qh16it">Here is how the different processors compose. Each arrow is a processor step, and they can be chained in a <code>RobotProcessorPipeline</code> or <code>PolicyProcessorPipeline</code>:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --> ┌─────────────────────────────────────────┐ | |
| Action Space │ Joint Space ←──IK──→ EE Space │ | |
| │ ForwardKinematicsJointsToEE │ | |
| │ InverseKinematicsEEToJoints │ | |
| └─────────────────────────────────────────┘ | |
| ┌─────────────────────────────────────────┐ | |
| Representation │ Absolute ←────→ Relative │ | |
| │ RelativeActionsProcessorStep (pre) │ | |
| │ AbsoluteActionsProcessorStep (post) │ | |
| └─────────────────────────────────────────┘ | |
| ┌─────────────────────────────────────────┐ | |
| Normalization │ Raw ←────→ Normalized │ | |
| │ NormalizerProcessorStep (pre) │ | |
| │ UnnormalizerProcessorStep (post) │ | |
| └─────────────────────────────────────────┘<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1tov15e">A typical training preprocessor might chain: <code>raw absolute joint actions → relative → normalize</code>. A typical inference postprocessor: <code>unnormalize → absolute → (optionally IK to joints)</code>.</p> <h2 class="relative group"><a id="references" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#references"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>References</span></h2> <ul data-svelte-h="svelte-jafhip"><li><a href="https://arxiv.org/abs/2402.10329" rel="nofollow">Universal Manipulation Interface (UMI)</a> - Chi et al., 2024. Defines the relative trajectory action representation and compares it with absolute and delta actions.</li> <li><a href="./introduction_processors">Introduction to Processors</a> - How processor pipelines work in LeRobot.</li> <li><a href="https://github.com/huggingface/lerobot/tree/main/examples/so100_to_so100_EE" rel="nofollow"><code>examples/so100_to_so100_EE/</code></a> - Complete example of recording and evaluating with EE-space actions.</li></ul> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/lerobot/blob/main/docs/source/action_representations.mdx" target="_blank"><svg class="mr-1" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M31,16l-7,7l-1.41-1.41L28.17,16l-5.58-5.59L24,9l7,7z"></path><path d="M1,16l7-7l1.41,1.41L3.83,16l5.58,5.59L8,23l-7-7z"></path><path d="M12.419,25.484L17.639,6.552l1.932,0.518L14.351,26.002z"></path></svg> <span data-svelte-h="svelte-zjs2n5"><span class="underline">Update</span> on GitHub</span></a> <p></p> | |
| <script> | |
| { | |
| __sveltekit_9kza6s = { | |
| assets: "/docs/lerobot/pr_3313/en", | |
| base: "/docs/lerobot/pr_3313/en", | |
| env: {} | |
| }; | |
| const element = document.currentScript.parentElement; | |
| const data = [null,null]; | |
| Promise.all([ | |
| import("/docs/lerobot/pr_3313/en/_app/immutable/entry/start.d3f1c0f3.js"), | |
| import("/docs/lerobot/pr_3313/en/_app/immutable/entry/app.04bb7687.js") | |
| ]).then(([kit, app]) => { | |
| kit.start(app, element, { | |
| node_ids: [0, 3], | |
| data, | |
| form: null, | |
| error: null | |
| }); | |
| }); | |
| } | |
| </script> | |
Xet Storage Details
- Size:
- 60.6 kB
- Xet hash:
- 9dd8b0b386f5003d355cebf535ee06e886018f8522d83286832b4444fda9e1e3
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.