Upload data and config files

data/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:78d3ea7fc4c8fc294575913c669e0116616f2a73069c0947847c51b5ae639473
+size 872841

toolshed_config.yaml

@@ -0,0 +1,470 @@
+tools:
+- class_name: toolshed.integration.verl.ToolshedMethodTool
+  config:
+    type: native
+    router_name: toolshed_router
+    namespace: toolshed
+    function_name: robot.capture_image
+    timeout: 30
+  tool_schema:
+    type: function
+    function:
+      name: robot.capture_image
+      description: 'Capture an RGB image from the robot''s camera showing the current
+        scene.
+
+        Text output: Image dimensions and capture status.
+
+        Image output: RGB image from robot camera.'
+      parameters:
+        type: object
+        properties: {}
+        required: []
+- class_name: toolshed.integration.verl.ToolshedMethodTool
+  config:
+    type: native
+    router_name: toolshed_router
+    namespace: toolshed
+    function_name: robot.execute_grasp
+    timeout: 30
+  tool_schema:
+    type: function
+    function:
+      name: robot.execute_grasp
+      description: "Execute a grasp by moving the robot to the specified pose via\
+        \ a pre-grasp point, and closing the gripper.\nText output: Status of the\
+        \ grasp execution.\n\nArgs:\n grasp_pose: 4\xD74 transformation matrix (list\
+        \ or numpy array) representing the grasp pose\n in the robot's camera frame\
+        \ (OpenCV convention).\n\nReturns:\n ToolResult: value dict with ``success``\
+        \ (bool) and ``execution_time_s`` (float)."
+      parameters:
+        type: object
+        properties:
+          grasp_pose:
+            type: string
+            description: Parameter grasp_pose
+        required:
+        - grasp_pose
+- class_name: toolshed.integration.verl.ToolshedMethodTool
+  config:
+    type: native
+    router_name: toolshed_router
+    namespace: toolshed
+    function_name: robot.get_depth
+    timeout: 30
+  tool_schema:
+    type: function
+    function:
+      name: robot.get_depth
+      description: 'Retrieve depth map from the robot''s depth sensor.
+
+        Text output: Summary of depth data including image dimensions, focal length,
+        and depth statistics.'
+      parameters:
+        type: object
+        properties: {}
+        required: []
+- class_name: toolshed.integration.verl.ToolshedMethodTool
+  config:
+    type: native
+    router_name: toolshed_router
+    namespace: toolshed
+    function_name: robot.get_depth_with_pointcloud
+    timeout: 30
+  tool_schema:
+    type: function
+    function:
+      name: robot.get_depth_with_pointcloud
+      description: 'Retrieve depth map from robot''s depth sensor and generate 3D
+        point cloud.
+
+        Text output: Summary of depth data and point cloud generation including dimensions,
+        focal length, depth statistics, and point cloud size.'
+      parameters:
+        type: object
+        properties: {}
+        required: []
+- class_name: toolshed.integration.verl.ToolshedMethodTool
+  config:
+    type: native
+    router_name: toolshed_router
+    namespace: toolshed
+    function_name: robot.place_object_at_2d_location
+    timeout: 30
+  tool_schema:
+    type: function
+    function:
+      name: robot.place_object_at_2d_location
+      description: "Simulate placing object at 2D location (always succeeds).\nText\
+        \ output: Confirmation that placement was successful.\n\nArgs:\n placement_point_2d:\
+        \ 2D normalized image coordinate [x, y] in the range [0, 1]\n where the object\
+        \ should be placed.\n\nReturns:\n ToolResult: value dict with ``success``\
+        \ (bool) and ``execution_time_s`` (float)."
+      parameters:
+        type: object
+        properties:
+          placement_point_2d:
+            type: string
+            description: Parameter placement_point_2d
+        required:
+        - placement_point_2d
+- class_name: toolshed.integration.verl.ToolshedMethodTool
+  config:
+    type: native
+    router_name: toolshed_router
+    namespace: toolshed
+    function_name: robot.place_object_at_3d_location
+    timeout: 30
+  tool_schema:
+    type: function
+    function:
+      name: robot.place_object_at_3d_location
+      description: "Simulate placing object at 3D location (always succeeds).\nText\
+        \ output: Confirmation that placement was successful.\n\nArgs:\n placement_point_3d:\
+        \ 3D point [x, y, z] in the robot's camera frame (list or numpy array)\n where\
+        \ the object should be placed.\n\nReturns:\n ToolResult: value dict with ``success``\
+        \ (bool) and ``execution_time_s`` (float)."
+      parameters:
+        type: object
+        properties:
+          placement_point_3d:
+            type: string
+            description: Parameter placement_point_3d
+        required:
+        - placement_point_3d
+- class_name: toolshed.integration.verl.ToolshedMethodTool
+  config:
+    type: native
+    router_name: toolshed_router
+    namespace: toolshed
+    function_name: vision_ops.index_at
+    timeout: 30
+  tool_schema:
+    type: function
+    function:
+      name: vision_ops.index_at
+      description: 'Get the pixel value in the numpy ndarray `data` at the given normalized
+        coordinates (u,v).
+
+        Note the input arguments are called `data`, `u`, `v`.
+
+
+        Text output: Information about the pixel value at the given coordinates.'
+      parameters:
+        type: object
+        properties:
+          data:
+            type: string
+            description: Numpy ndarray of shape (H, W) or (H, W, C), or PIL Image
+          u:
+            type: number
+            description: Normalized x-coordinate in [0, 1]
+          v:
+            type: number
+            description: Normalized y-coordinate in [0, 1]
+        required:
+        - data
+        - u
+        - v
+- class_name: toolshed.integration.verl.ToolshedMethodTool
+  config:
+    type: native
+    router_name: toolshed_router
+    namespace: toolshed
+    function_name: bounding_box.compute_bbox
+    timeout: 30
+  tool_schema:
+    type: function
+    function:
+      name: bounding_box.compute_bbox
+      description: 'Compute an oriented bounding box for a masked subset of a point
+        cloud.
+
+        Text output: Summary containing number of input points, the point coordinates
+        in 3d and 2d,
+
+        mask shape, box extents, and edges.'
+      parameters:
+        type: object
+        properties:
+          point_cloud:
+            type: string
+            description: scene point cloud, np.ndarray of shape (N, 3) with float
+              values.
+          mask:
+            type: string
+            description: np.ndarray of shape (H, W) with boolean values segmenting
+              the target object. Dimensions are used for camera projection.
+          focal_length_px:
+            type: number
+            description: Camera focal length in pixels (square pixels assumed).
+        required:
+        - point_cloud
+        - mask
+        - focal_length_px
+- class_name: toolshed.integration.verl.ToolshedMethodTool
+  config:
+    type: native
+    router_name: toolshed_router
+    namespace: toolshed
+    function_name: sam2.segment_from_point
+    timeout: 30
+  tool_schema:
+    type: function
+    function:
+      name: sam2.segment_from_point
+      description: 'Segment the object at normalized coordinates.
+
+        Text output: Summary of segmentation including mask dimensions and IoU (Intersection
+        over Union) confidence score.'
+      parameters:
+        type: object
+        properties:
+          x:
+            type: number
+            description: X-coordinate in normalized range [0, 1].
+          y:
+            type: number
+            description: Y-coordinate in normalized range [0, 1].
+          image_index:
+            type: integer
+            description: 0-based index of an image already present in the conversation.
+              Use this to reference an inline image by its order (0 = first, 1 = second,
+              ...).
+        required:
+        - x
+        - y
+        - image_index
+- class_name: toolshed.integration.verl.ToolshedMethodTool
+  config:
+    type: native
+    router_name: toolshed_router
+    namespace: toolshed
+    function_name: sam2.segment_from_points
+    timeout: 30
+  tool_schema:
+    type: function
+    function:
+      name: sam2.segment_from_points
+      description: 'Segment an object using multiple normalized coordinates.
+
+        Text output: Summary of segmentation including mask dimensions and IoU scores
+        for candidate masks.'
+      parameters:
+        type: object
+        properties:
+          points:
+            type: string
+            description: Sequence of (x, y) normalized coordinates in [0, 1] range.
+          image_index:
+            type: integer
+            description: 0-based index of an image already present in the conversation.
+              Use this to reference an inline image by its order (0 = first, 1 = second,
+              ...).
+        required:
+        - points
+        - image_index
+- class_name: toolshed.integration.verl.ToolshedMethodTool
+  config:
+    type: native
+    router_name: toolshed_router
+    namespace: toolshed
+    function_name: depth_estimator.estimate_depth
+    timeout: 30
+  tool_schema:
+    type: function
+    function:
+      name: depth_estimator.estimate_depth
+      description: 'Estimate depth map from a single image.
+
+        Text output: Summary of depth estimation including image dimensions, focal
+        length, and depth statistics.'
+      parameters:
+        type: object
+        properties:
+          image_index:
+            type: integer
+            description: 0-based index of an image already present in the conversation.
+              Use this to reference an inline image by its order (0 = first, 1 = second,
+              ...).
+        required:
+        - image_index
+- class_name: toolshed.integration.verl.ToolshedMethodTool
+  config:
+    type: native
+    router_name: toolshed_router
+    namespace: toolshed
+    function_name: depth_estimator.estimate_depth_with_pointcloud
+    timeout: 30
+  tool_schema:
+    type: function
+    function:
+      name: depth_estimator.estimate_depth_with_pointcloud
+      description: 'Estimate depth and generate 3D point cloud from a single image.
+
+        Text output: Summary of depth estimation and point cloud generation including
+        dimensions, focal length, depth statistics, and point cloud size.'
+      parameters:
+        type: object
+        properties:
+          image_index:
+            type: integer
+            description: 0-based index of an image already present in the conversation.
+              Use this to reference an inline image by its order (0 = first, 1 = second,
+              ...).
+        required:
+        - image_index
+- class_name: toolshed.integration.verl.ToolshedMethodTool
+  config:
+    type: native
+    router_name: toolshed_router
+    namespace: toolshed
+    function_name: grasp_generator.compute_grasp
+    timeout: 30
+  tool_schema:
+    type: function
+    function:
+      name: grasp_generator.compute_grasp
+      description: 'Generate a single grasp pose for a masked subset of a point cloud.
+
+        Text output: Confidence score, number of masked points used, projected 2D
+        gripper points in normalize image coordinates.'
+      parameters:
+        type: object
+        properties:
+          point_cloud:
+            type: string
+            description: Nx3 numpy float array. Full scene point cloud.
+          mask:
+            type: string
+            description: Boolean mask aligning with *image*; accepts ndarray. Indicates
+              object points.
+          focal_length_px:
+            type: number
+            description: Camera focal length in pixels (square pixels assumed).
+          image_index:
+            type: integer
+            description: 0-based index of an image already present in the conversation.
+              Use this to reference an inline image by its order (0 = first, 1 = second,
+              ...).
+        required:
+        - point_cloud
+        - mask
+        - focal_length_px
+        - image_index
+- class_name: toolshed.integration.verl.ToolshedMethodTool
+  config:
+    type: native
+    router_name: toolshed_router
+    namespace: toolshed
+    function_name: roborefer.detect_all
+    timeout: 30
+  tool_schema:
+    type: function
+    function:
+      name: roborefer.detect_all
+      description: 'Detect *all* instances of *obj_name* in *image*.
+
+        Text output: List of point coordinates for the detected objects, in normalized
+        pixel space in range [0, 1].'
+      parameters:
+        type: object
+        properties:
+          obj_name:
+            type: string
+            description: Name or description of the object to detect.
+          image_index:
+            type: integer
+            description: 0-based index of an image already present in the conversation.
+              Use this to reference an inline image by its order (0 = first, 1 = second,
+              ...).
+        required:
+        - obj_name
+        - image_index
+- class_name: toolshed.integration.verl.ToolshedMethodTool
+  config:
+    type: native
+    router_name: toolshed_router
+    namespace: toolshed
+    function_name: roborefer.detect_one
+    timeout: 30
+  tool_schema:
+    type: function
+    function:
+      name: roborefer.detect_one
+      description: 'Detect *one* instance of *obj_name* in *image*.
+
+        Text output: coordinates of a single point for the first instance of the object,
+        in normalized pixel space in range [0, 1].'
+      parameters:
+        type: object
+        properties:
+          obj_name:
+            type: string
+            description: Parameter obj_name
+          image_index:
+            type: integer
+            description: 0-based index of an image already present in the conversation.
+              Use this to reference an inline image by its order (0 = first, 1 = second,
+              ...).
+        required:
+        - obj_name
+        - image_index
+- class_name: toolshed.integration.verl.ToolshedMethodTool
+  config:
+    type: native
+    router_name: toolshed_router
+    namespace: toolshed
+    function_name: vlm.detect_all
+    timeout: 30
+  tool_schema:
+    type: function
+    function:
+      name: vlm.detect_all
+      description: 'Detect *all* instances of *obj_name* in *image*.
+
+        Text output: List of point coordinates for the detected objects, in normalized
+        pixel space in range [0, 1].'
+      parameters:
+        type: object
+        properties:
+          obj_name:
+            type: string
+            description: Name or description of the object to detect.
+          image_index:
+            type: integer
+            description: 0-based index of an image already present in the conversation.
+              Use this to reference an inline image by its order (0 = first, 1 = second,
+              ...).
+        required:
+        - obj_name
+        - image_index
+- class_name: toolshed.integration.verl.ToolshedMethodTool
+  config:
+    type: native
+    router_name: toolshed_router
+    namespace: toolshed
+    function_name: vlm.detect_one
+    timeout: 30
+  tool_schema:
+    type: function
+    function:
+      name: vlm.detect_one
+      description: 'Detect *one* instance of *obj_name* in *image*.
+
+        Text output: coordinates of a single point for the first instance of the object,
+        in normalized pixel space in range [0, 1].'
+      parameters:
+        type: object
+        properties:
+          obj_name:
+            type: string
+            description: Parameter obj_name
+          image_index:
+            type: integer
+            description: 0-based index of an image already present in the conversation.
+              Use this to reference an inline image by its order (0 = first, 1 = second,
+              ...).
+        required:
+        - obj_name
+        - image_index