assawayut

assawayut/blendmpc

Composable blends of MPC and RL for robotics: residual RL, learned terminal costs, policy warm starts, and MPC distillation over Gymnasium + Crocoddyl/acados, benchmarked on a walking Go2

1 stars 0 forks 1 watchers Python MIT License

acados crocoddyl gymnasium legged-robots model-predictive-control mpc mujoco quadruped reinforcement-learning robotics

View on GitHub Website

5 Open Issues Need Help Last updated: Sep 12, 2026

Open Issues Need Help

View All on GitHub

Colab notebook for the residual pendulum example 7 days ago

documentation good first issue

assawayut/blendmpc

1

Composable blends of MPC and RL for robotics: residual RL, learned terminal costs, policy warm starts, and MPC distillation over Gymnasium + Crocoddyl/acados, benchmarked on a walking Go2

Python

#acados#crocoddyl#gymnasium#legged-robots#model-predictive-control#mpc#mujoco#quadruped#reinforcement-learning#robotics

API reference in the docs with mkdocstrings 7 days ago

documentation good first issue

assawayut/blendmpc

1

Composable blends of MPC and RL for robotics: residual RL, learned terminal costs, policy warm starts, and MPC distillation over Gymnasium + Crocoddyl/acados, benchmarked on a walking Go2

Python

#acados#crocoddyl#gymnasium#legged-robots#model-predictive-control#mpc#mujoco#quadruped#reinforcement-learning#robotics

Scale trot swing timing with commanded speed (envelope ends ~0.4 m/s) 7 days ago

enhancement good first issue

assawayut/blendmpc

1

Composable blends of MPC and RL for robotics: residual RL, learned terminal costs, policy warm starts, and MPC distillation over Gymnasium + Crocoddyl/acados, benchmarked on a walking Go2

Python

#acados#crocoddyl#gymnasium#legged-robots#model-predictive-control#mpc#mujoco#quadruped#reinforcement-learning#robotics

Lateral velocity and yaw-rate commands for the Go2 trot 7 days ago

enhancement help wanted good first issue

assawayut/blendmpc

1

Composable blends of MPC and RL for robotics: residual RL, learned terminal costs, policy warm starts, and MPC distillation over Gymnasium + Crocoddyl/acados, benchmarked on a walking Go2

Python

#acados#crocoddyl#gymnasium#legged-robots#model-predictive-control#mpc#mujoco#quadruped#reinforcement-learning#robotics

Signed symmetry-breaking cold init for the pendulum (seed 113 hangs) 7 days ago

enhancement good first issue

assawayut/blendmpc

1

Composable blends of MPC and RL for robotics: residual RL, learned terminal costs, policy warm starts, and MPC distillation over Gymnasium + Crocoddyl/acados, benchmarked on a walking Go2

Python

#acados#crocoddyl#gymnasium#legged-robots#model-predictive-control#mpc#mujoco#quadruped#reinforcement-learning#robotics