DecAP: Decaying Action Priors for Accelarated Learning of Torque-based Legged Locomotion Policies

The repositiry contains the research code for the IROS paper : DecAP.

@article{Sood2023DecAPDA,
  title={DecAP: Decaying Action Priors for Accelerated Learning of Torque-Based Legged Locomotion Policies},
  author={Shivam Sood and Ge Sun and Peizhuo Li and Guillaume Sartoretti},
  journal={ArXiv},
  year={2023},
  volume={abs/2310.05714},
  url={https://api.semanticscholar.org/CorpusID:263830010}
}

Affiliation: MARMoT Lab, NUS
Contact: [email protected]

---

Follow leggged_gym repo for setting up the initial environment.

Running the trained examples

The trained policies are in the logs folder. Each folder has torque policies trained with DecAP + Imitation and Imitation alone. The logs are named decap_[reward_scale_value] and imi_[reward_scale_value].

To run the torque policy set control_type = T, action_scale = 8.0 in the robot config file. Then run the following command:

cd decaying_action_priors
python legged_gym/scripts/play.py --task=go1_flat --load_run=decap_0.75x

Training the torque policy

Set the action_scale = 8.0

Imitation Rewards:

• Joint Angles
• End-effector position
• Foot height
• Base height

1. With imitation and DecAP: Set the control_type
   

   • For Go1 (Unitree): T_ref_decay
     
   • For Yuna (Hebi-Daisy) : T_vanish_yuna
     
   • For Cassie (Agility) : T_vanish_humanoid
     

   Uncomment the imitation rewards mentioned above.
   Imitation Reward scales tested: 0.75, 1.5, 7.5, 15.0

2. With just imitation data:
   Set the control_type = T

   Uncomment the imitation rewards mentioned

Getting the imitation data:

Train a position policy/ use any position policy or Optimal Controller to get imitation data

Trained position policies are in the logs folder and can also be used to collect the imitation data. For running these position polcies set action_scale = 0.25 and control_type = P. The imitation data folder contains the data used.
The imitation rewards are defined in the respective robot files in the envs folder. Decaying Action Priors are set in compute_torques function in legged_robot.py