This repository is about trash collection using RL on quadruped robots, with sim-to-sim and sim-to-real scripts. It trains in sequence two different policies, one for locomotion and a second one for manipulation, able to reach desired end effector poses. In alternative, an inverse kinematic using mink can be used to reach the target.
Features:
- Locomotion policy able to adjust pose and carry a manipulator
- Manipulation policy able to reach desired end effector goal while coordinating the pose of the quadruped (work in progress - pull request well accepted!)
- Whole-body Inverse Kinematics with reduced model using mink
- Sim-to-Sim in Mujoco
- Sim-to-Real in ROS2 compatible with our public low-level robot's hal for Unitree robots (only z1 for now available - not aliengo) unitree_ros2_dls
A list of robots available and envs are described below:
| Robot Model | Environment Name (ID) |
|---|---|
| Aliengo | Locomotion-Aliengo-Flat, Locomotion-Aliengo-Rough-Blind, Locomotion-Aliengo-Rough-Vision |
| Arm with Z1 | Manipulation-Aliengo-Flat, Manipulation-Aliengo-Rough-Blind, Manipulation-Aliengo-Rough-Vision |
If you want only to deploy a trained policy on your robot, continue on README_DEPLOY otherwise on README_TRAIN.
If you find the work useful, please consider citing:
BinWalker: Development and Field Evaluation of a Quadruped Manipulator Platform for Sustainable Litter Collection
@article{turrisi26littercollection,
author = {Giulio Turrisi and Angelo Bratta and Giovanni Minelli and Gabriel Fischer Abati and Amir H. Rad and João Carlos Virgolino Soares and Claudio Semini},
title = {BinWalker: Development and Field Evaluation of a Quadruped Manipulator Platform for Sustainable Litter Collection},
journal = {arXiv},
year = {2026}
}
This repository is maintained by Giulio Turrisi.