Commit 3bda6f0
authored
K1 RL booster walking (#2132)
* Add boosters T1 walking checkpoint
* Implement RL walking inference
Implemented as in [BoosterRobotics/booster_gym](https://github.com/BoosterRobotics/booster_gym?tab=readme-ov-file) `deploy/`.
* Compute `LowCommand` from inference outputs
Add ".onnx" files to lfs
* Add ' WalkingInferenceInputs` as `AdditionalOutput`
* Initially send preparation joint positions before using inference outputs
* Add correct K1 motor command parameters
* Publish low command as MainOutput
* Intialize target joints with prepare motor command
* Use sum of default positions and inference outputs
* Normalize joint position and velocities in input
* Correctly compute gravity vector
* Cleanup and simplification
* Add parameter for walk command
* Joint target smoothing
* Make prepare pose stand on its own
* Impl dummy `TimeInterface` for imagine and replayer
* `Joints<f32>` -> `Joints<MotorState>`
* Move walking inference to separate crate
* Directly use inference outputs instead of prepare pose
* Address review comments
* Increment `gait_progress` with `last_cycle_duration`
* Address review comments II
* Refactor: move state into `walking_inference`
* Rename `rl_walking` -> `booster_walking`
* Set `ort` log level to warn1 parent 8b29531 commit 3bda6f0
28 files changed
Lines changed: 1153 additions & 70 deletions
File tree
- crates
- booster
- src
- control
- src
- motion
- hulk_imagine
- src
- hulk_manifest/src
- hulk_mujoco
- src
- hulk_replayer
- src
- types
- src
- walking_inference
- src
- etc
- neural_networks
- parameters
- tools/parameter_tester
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
7 | 7 | | |
8 | 8 | | |
9 | 9 | | |
| 10 | + | |
0 commit comments