Reinforcement Learning-based adaptive control for legged robots

Train dual_history policy

python legged_gym/scripts/train_dual.py --headless --task=a1 --sim_device=cuda:0 --rl_device=cuda:0

Evaluation

python legged_gym/scripts/eval.py --headless --task=a1 --sim_device=cuda:0 --rl_device=cuda:0

Radar Ploting

python adaptivity.py <a.pt> <b.pt>

It would be better add some args like --net_type=, --hist_design=,

Metrics Description

• Command velocity test
  • Fixed lin_vel_x
  • Fixed lin_vel_y
• Friction test
• Add Mass test