How langchengg/rl-agibot-x2 works
AgiBot X2 Ground Recovery is a MuJoCo/Gymnasium supine-recovery controller, with a ROS 2 runtime using the same frozen controller bundle. The primary workflow loads controller inputs, simulates a recovery, and judges stable standing and safety constraints; PPO training and reference-motion search support controller development. Evaluation, strict reproduction, diagnostics, and ROS telemetry are distinct supporting capabilities. The README documents these capabilities, but sampled source excerpts are sparse, so uncertain internal call relationships are omitted.
Main entry points: env.py Recovery environment · recovery_node.py Recovery ROS node · model.py Robot model · train.py PPO training
Components 19
Recovery runtime
- Recovery environment
src/x2_recovery/x2_recovery/env.py - Robot model
src/x2_recovery/x2_recovery/model.py - Pose reset
src/x2_recovery/x2_recovery/reset.py - Scripted baseline
src/x2_recovery/x2_recovery/baseline.py - Success tracking
src/x2_recovery/x2_recovery/success.py
Controller development
- PPO training
src/x2_recovery/x2_recovery/train.py - Reference search
src/x2_recovery/x2_recovery/motion_search.py
Evaluation and audit
- Strict reproduction
src/x2_recovery/x2_recovery/reproduce.py - Policy evaluation
src/x2_recovery/x2_recovery/evaluate.py - Diagnostics
src/x2_recovery/x2_recovery/diagnostics.py - Model audit
src/x2_recovery/x2_recovery/model_audit.py - Constraint audit
src/x2_recovery/x2_recovery/constraint_audit.py - Simulation validation
src/x2_recovery/x2_recovery/simulation_validation.py - Success validation
src/x2_recovery/x2_recovery/success_validation.py
ROS integration
- Recovery ROS node
src/x2_recovery/x2_recovery/recovery_node.py - Telemetry node
src/x2_recovery/x2_recovery/telemetry_node.py - ROS launch
src/x2_recovery/launch/recovery.launch.py
Other components
- Operator
- X2 robot assets
Connections 21
- Operator → Strict reproduction: requests replay
- Strict reproduction → Policy evaluation: runs episode
- Policy evaluation → Recovery environment: simulates recovery
- Recovery environment → Robot model: uses model
- Robot model → X2 robot assets: loads assets
- Recovery environment → Pose reset: resets pose
- Recovery environment → Success tracking: checks standing
- Operator → PPO training: trains policy
- PPO training → Recovery environment: trains in
- Operator → Reference search: searches references
- Reference search → Recovery environment: tests candidates
- Operator → Recovery ROS node: starts recovery
- ROS launch → Recovery ROS node: launches
- Recovery ROS node → Recovery environment: runs controller
- Recovery ROS node → Scripted baseline: selects debug mode
- Telemetry node → Recovery environment: observes runtime
- Operator → Diagnostics: requests analysis
- Operator → Model audit: audits model
- Operator → Constraint audit: audits constraints
- Operator → Simulation validation: validates simulation
- Operator → Success validation: validates success