deep-rl-course/pr_587/en/unit7/multi-agent-setting.html

<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Designing Multi-Agents systems&quot;,&quot;local&quot;:&quot;designing-multi-agents-systems&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Decentralized system&quot;,&quot;local&quot;:&quot;decentralized-system&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Centralized approach&quot;,&quot;local&quot;:&quot;centralized-approach&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}">
		<link href="/docs/deep-rl-course/pr_587/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
		<link rel="modulepreload" href="/docs/deep-rl-course/pr_587/en/_app/immutable/entry/start.7c4c7929.js">
		<link rel="modulepreload" href="/docs/deep-rl-course/pr_587/en/_app/immutable/chunks/scheduler.37c15a92.js">
		<link rel="modulepreload" href="/docs/deep-rl-course/pr_587/en/_app/immutable/chunks/singletons.945d9b5d.js">
		<link rel="modulepreload" href="/docs/deep-rl-course/pr_587/en/_app/immutable/chunks/index.18351ede.js">
		<link rel="modulepreload" href="/docs/deep-rl-course/pr_587/en/_app/immutable/chunks/paths.9ffa1e06.js">
		<link rel="modulepreload" href="/docs/deep-rl-course/pr_587/en/_app/immutable/entry/app.d2a43a08.js">
		<link rel="modulepreload" href="/docs/deep-rl-course/pr_587/en/_app/immutable/chunks/index.7cb9c9b8.js">
		<link rel="modulepreload" href="/docs/deep-rl-course/pr_587/en/_app/immutable/nodes/0.e67798fa.js">
		<link rel="modulepreload" href="/docs/deep-rl-course/pr_587/en/_app/immutable/chunks/each.e59479a4.js">
		<link rel="modulepreload" href="/docs/deep-rl-course/pr_587/en/_app/immutable/nodes/76.50efc908.js">
		<link rel="modulepreload" href="/docs/deep-rl-course/pr_587/en/_app/immutable/chunks/Youtube.8666c400.js">
		<link rel="modulepreload" href="/docs/deep-rl-course/pr_587/en/_app/immutable/chunks/EditOnGithub.858acfec.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Designing Multi-Agents systems&quot;,&quot;local&quot;:&quot;designing-multi-agents-systems&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Decentralized system&quot;,&quot;local&quot;:&quot;decentralized-system&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Centralized approach&quot;,&quot;local&quot;:&quot;centralized-approach&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"><!-- HEAD_svelte-u9bgzb_END -->      <p></p>  <h1 class="relative group"><a id="designing-multi-agents-systems" 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="#designing-multi-agents-systems"><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>Designing Multi-Agents systems</span></h1> <p data-svelte-h="svelte-s0u6gv">For this section, you’re going to watch this excellent introduction to multi-agents made by <a href="https://www.youtube.com/channel/UCq0imsn84ShAe9PBOFnoIrg">Brian Douglas </a>.</p> <iframe class="w-full xl:w-4/6 h-80" src="https://www.youtube-nocookie.com/embed/qgb0gyrpiGk" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> <p data-svelte-h="svelte-12o0rkz">In this video, Brian talked about how to design multi-agent systems. He specifically took a multi-agents system of vacuum cleaners and asked: <strong>how can can cooperate with each other</strong>?</p> <p data-svelte-h="svelte-h4778x">We have two solutions to design this multi-agent reinforcement learning system (MARL).</p>  <h2 class="relative group"><a id="decentralized-system" 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="#decentralized-system"><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>Decentralized system</span></h2> <figure data-svelte-h="svelte-18z4kzt"><img src="https://huggingface.co/datasets/huggingface-deep-rl-course/course-images/resolve/main/en/unit10/decentralized.png" alt="Decentralized"> <figcaption>Source: <a href="https://www.youtube.com/watch?v=qgb0gyrpiGk">Introduction to Multi-Agent Reinforcement Learning</a></figcaption></figure> <p data-svelte-h="svelte-ljzs98">In decentralized learning, <strong>each agent is trained independently from the others</strong>. In the example given, each vacuum learns to clean as many places as it can <strong>without caring about what other vacuums (agents) are doing</strong>.</p> <p data-svelte-h="svelte-3o4vn0">The benefit is that <strong>since no information is shared between agents, these vacuums can be designed and trained like we train single agents</strong>.</p> <p data-svelte-h="svelte-c8dx7v">The idea here is that <strong>our training agent will consider other agents as part of the environment dynamics</strong>. Not as agents.</p> <p data-svelte-h="svelte-z6wvuu">However, the big drawback of this technique is that it will <strong>make the environment non-stationary</strong> since the underlying Markov decision process changes over time as other agents are also interacting in the environment.
And this is problematic for many Reinforcement Learning algorithms <strong>that can’t reach a global optimum with a non-stationary environment</strong>.</p>  <h2 class="relative group"><a id="centralized-approach" 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="#centralized-approach"><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>Centralized approach</span></h2> <figure data-svelte-h="svelte-ltt229"><img src="https://huggingface.co/datasets/huggingface-deep-rl-course/course-images/resolve/main/en/unit10/centralized.png" alt="Centralized"> <figcaption>Source: <a href="https://www.youtube.com/watch?v=qgb0gyrpiGk">Introduction to Multi-Agent Reinforcement Learning</a></figcaption></figure> <p data-svelte-h="svelte-ypcoh6">In this architecture, <strong>we have a high-level process that collects agents’ experiences</strong>: the experience buffer. And we’ll use these experiences <strong>to learn a common policy</strong>.</p> <p data-svelte-h="svelte-1owod73">For instance, in the vacuum cleaner example, the observation will be:</p> <ul data-svelte-h="svelte-dln72i"><li>The coverage map of the vacuums.</li> <li>The position of all the vacuums.</li></ul> <p data-svelte-h="svelte-14wt5x8">We use that collective experience <strong>to train a policy that will move all three robots in the most beneficial way as a whole</strong>. So each robot is learning from their common experience.
We now have a stationary environment since all the agents are treated as a larger entity, and they know the change of other agents’ policies (since it’s the same as theirs).</p> <p data-svelte-h="svelte-1uwte4">If we recap:</p> <ul data-svelte-h="svelte-1h1njmy"><li><p>In a <em>decentralized approach</em>, we <strong>treat all agents independently without considering the existence of the other agents.</strong></p> <ul><li>In this case, all agents <strong>consider others agents as part of the environment</strong>.</li> <li><strong>It’s a non-stationarity environment condition</strong>, so has no guarantee of convergence.</li></ul></li> <li><p>In a <em>centralized approach</em>:</p> <ul><li>A <strong>single policy is learned from all the agents</strong>.</li> <li>Takes as input the present state of an environment and the policy outputs joint actions.</li> <li>The reward is global.</li></ul></li></ul> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/deep-rl-class/blob/main/units/en/unit7/multi-agent-setting.mdx" target="_blank"><span data-svelte-h="svelte-1kd6by1">&lt;</span> <span data-svelte-h="svelte-x0xyl0">&gt;</span> <span data-svelte-h="svelte-1dajgef"><span class="underline ml-1.5">Update</span> on GitHub</span></a>  <p></p> 
			
			<script>
				{
					__sveltekit_18oo4fq = {
						assets: "/docs/deep-rl-course/pr_587/en",
						base: "/docs/deep-rl-course/pr_587/en",
						env: {}
					};

					const element = document.currentScript.parentElement;

					const data = [null,null];

					Promise.all([
						import("/docs/deep-rl-course/pr_587/en/_app/immutable/entry/start.7c4c7929.js"),
						import("/docs/deep-rl-course/pr_587/en/_app/immutable/entry/app.d2a43a08.js")
					]).then(([kit, app]) => {
						kit.start(app, element, {
							node_ids: [0, 76],
							data,
							form: null,
							error: null
						});
					});
				}
			</script>