Skip to main content
Policy-defined skills are neural network policies that run end-to-end—taking sensor input and directly outputting robot actions. Unlike code-defined skills where you write explicit logic, policy-defined skills learn behaviors from demonstration. This approach works well for manipulation tasks where the variability of real-world scenarios makes explicit programming impractical.

​
Two Types of Policies

Type Definition Use Case Learned Neural network trained on demonstrations Tasks requiring visual adaptation Replay Recorded action sequence Repeatable, fixed motions

​
Learned Skills (ACT Policy)

Learned skills use Action Chunking with Transformers (ACT)—a neural network architecture that takes camera images and joint positions as input, and outputs action sequences.

​
Metadata Structure

{
    "name": "pick_socks",
    "type": "learned",
    "guidelines": "Use when you need to pick up socks from the floor",
    "execution": {
        "model_type": "act_policy",
        "checkpoint": "act_policy_step_135000.pth",
        "stats_file": "dataset_stats.pt",
        "action_dim": 10,
        "duration": 45.0,
        "start_pose": [-0.015, -0.399, 1.456, -1.135, -0.023, 0.833]
    }
}

​
Execution Flow

  1. Load Policy: Neural network checkpoint loaded into GPU memory
  2. Move to Start Pose: Arm moves to consistent initial configuration
  3. Inference Loop (25Hz): Every 40ms:
    • Capture images from both cameras
    • Read current joint positions
    • Run policy forward pass
    • Output: 6 joint commands + 2 base velocity commands
  4. Progress Monitoring: Early termination when progress > 95%
  5. End Pose: Optionally return to safe configuration

​
Key Characteristics

  • Reactive: Continuously adjusts based on visual feedback
  • Adaptive: Handles variation in object position, lighting, orientation
  • Coordinated: Can move arm and base simultaneously

​
Replay Skills

Replay skills play back pre-recorded action sequences. Simpler than learned skills, but deterministic and reliable.

​
Metadata Structure

{
    "name": "wave",
    "type": "replay",
    "guidelines": "Use when greeting someone or saying hello",
    "execution": {
        "model_type": "replay",
        "replay_file": "episode_0.h5",
        "replay_frequency": 50.0,
        "start_pose": [1.577, -0.6, 1.477, -0.738, 0.0, 0.0],
        "end_pose": [1.577, -0.6, 1.477, -0.738, 0.0, 0.0]
    }
}

​
Execution Flow

  1. Load Recording: H5 file containing timestamped actions
  2. Move to Start Pose: Arm moves to recorded initial position
  3. Playback (50Hz): Execute recorded actions in sequence
  4. End Pose: Return to specified configuration

​
Execution Pipeline

Both skill types are executed by the BehaviorServer: 
BASIC calls skill
       |
       v
PrimitiveExecutionActionServer
       |
       v (physical skill detected)
BehaviorServer.ExecuteBehavior
       |
       +-- Learned --> Load policy, run inference loop
       |
       +-- Replay ---> Load H5 file, playback loop
       |
       v
Robot hardware (/mars/arm/commands, /cmd_vel)

​
Creating New Skills

​
Learned Skill Workflow

  1. Collect Demonstrations: Teleoperate robot through task 10-100 times
  2. Train Policy: Run ACT training pipeline to produce checkpoint
  3. Create Metadata: Add metadata.json with execution parameters
  4. Deploy: Place in skills/<name>/ directory

​
Replay Skill Workflow

  1. Record Episode: Teleoperate through motion once, save to H5
  2. Create Metadata: Add metadata.json with replay parameters
  3. Deploy: Place in skills/<name>/ directory

​
Demonstration Quality

The quality of learned skills depends directly on demonstration quality: Good Demonstrations Poor Demonstrations Consistent starting positions Variable starting positions Smooth, deliberate motions Hesitant, jerky motions Varied object positions Always same position Include error recovery Only perfect executions

​
Skill Selection Guide

Scenario Recommended Type Task requires visual adaptation Learned Object position varies Learned Motion must be identical every time Replay Simple gesture (wave, point) Replay Faster development cycle Replay