How mint-sjtu/evo-rl works
Evo-RL is a LeRobot-based toolkit for real-world robot learning. Its documented workflow runs from hardware setup and demonstration collection through value-function training and inference, policy training, and closed-loop rollout for another data-collection round. The graph includes that workflow, shared policy/processor infrastructure, and notable policy and robot-runtime capabilities. Workflow links follow the README; source-index references support only the sampled internal links, so unsampled implementation details are not inferred.
Main entry points: lerobot_train.py Policy Training · lerobot_dataset.py Robot Dataset · robot.py Robot Hardware · factory.py Policy Factory
Components 25
Robot Data Collection
- Robot Hardware
src/lerobot/robots/robot.py - Teleoperation
src/lerobot/teleoperators/teleoperator.py - Camera Capture
src/lerobot/cameras/camera.py - Data Recording
src/lerobot/scripts/lerobot_record.py - Robot Dataset
src/lerobot/datasets/lerobot_dataset.py - Subtask Annotation
src/lerobot/data_processing/sarm_annotations/subtask_annotation.py
Value And Policy Learning
- Value Training
src/lerobot/scripts/lerobot_value_train.py - Value Model
src/lerobot/values/pistar06/modeling_pistar06.py - Value Inference
src/lerobot/scripts/lerobot_value_infer.py - Policy Training
src/lerobot/scripts/lerobot_train.py
Policy Runtime
- Policy Factory
src/lerobot/policies/factory.py - Policy Models
src/lerobot/policies/pretrained.py - Wall-X Policy
src/lerobot/policies/wall_x/modeling_wall_x.py - Qwen Vision Model
src/lerobot/policies/wall_x/qwen_model/qwen2_5_vl_moe.py - Realtime Control
src/lerobot/policies/rtc/modeling_rtc.py - Policy Evaluation
src/lerobot/scripts/lerobot_eval.py - Robot Teleoperation
src/lerobot/scripts/lerobot_teleoperate.py - Async Robot Client
src/lerobot/async_inference/robot_client.py
Shared LeRobot Foundation
- Data Processors
src/lerobot/processor/pipeline.py - Data Converters
src/lerobot/processor/converters.py - Policy Configuration
src/lerobot/configs/policies.py - Optimizers
src/lerobot/optim/optimizers.py - Schedulers
src/lerobot/optim/schedulers.py - Hub Utilities
src/lerobot/utils/hub.py
Other components
- Robot Operator
Connections 26
- Robot Operator → Teleoperation: operates
README.md - Teleoperation → Robot Hardware: controls
README.md - Camera Capture → Data Recording: captures frames
README.md - Robot Hardware → Data Recording: provides observations
README.md - Data Recording → Robot Dataset: records episodes
README.md - Robot Dataset → Value Training: trains from
README.md - Robot Dataset → Policy Training: trains from
README.md - Robot Dataset → Subtask Annotation: annotates
src/lerobot/data_processing/sarm_annotations/subtask_annotation.py - Value Training → Value Model: trains
README.md - Value Model → Value Inference: scores episodes
README.md - Value Inference → Policy Training: informs
README.md - Policy Training → Policy Factory: configures policy
src/lerobot/policies/factory.py - Policy Factory → Policy Configuration: loads configuration
src/lerobot/policies/factory.py - Policy Factory → Policy Models: creates model
src/lerobot/policies/factory.py - Policy Training → Optimizers: uses
- Policy Training → Schedulers: uses
- Wall-X Policy → Qwen Vision Model: uses vision model
src/lerobot/policies/wall_x/modeling_wall_x.py - Wall-X Policy → Realtime Control: supports realtime control
src/lerobot/policies/rtc/modeling_rtc.py - Policy Models → Hub Utilities: publishes models
src/lerobot/policies/pretrained.py - Policy Training → Data Processors: processes policy data
- Data Processors → Data Converters: uses converters
src/lerobot/processor/pipeline.py - Policy Evaluation → Policy Factory: loads policy
- Policy Evaluation → Robot Hardware: runs rollout
README.md - Policy Evaluation → Robot Dataset: records next round
README.md - Robot Teleoperation → Robot Hardware: drives hardware
README.md - Async Robot Client → Robot Hardware: sends actions
src/lerobot/async_inference/robot_client.py