Humanoid robots plan complex moves for picking carrying and kicking
DiPLO: Self-distilled Diffusion Planning for Loco-manipulation
Robotics
Summary
Humanoid robots have a hard time doing tasks like walking while holding and moving objects because it’s tricky to control their balance and movements together. The authors created DiPLO, a system that helps robots plan and perform these tasks by learning from human motions and improving through practice. They teach the robot a plan and a way to follow it, then refine the plan so it works better with how the robot moves. This leads to robots successfully doing tasks like picking up and placing objects almost all the time, even handling new objects and tasks without extra training.
What this means in practice
- •For robotics engineers: Generate coordinated motion plans for humanoid robots performing complex pick-and-place and manipulation tasks.
- •For industrial automation teams: Deploy robots capable of reliable object manipulation and locomotion in environments needing dynamic task execution without extensive retraining.
- •For robotics hardware developers: Integrate advanced planning and control techniques to improve motion execution on legged robot platforms like the Unitree G1 without real-world fine-tuning.$Commercial implications: Enables commercial robot products with enhanced autonomous manipulation capabilities on hardware platforms.
Authors
Shafeef Omar, Sebastian Sanokowski, Elham Daneshmand, Dian Yu, Daniel Cremers, Cordelia Schmid, Majid Khadiv
Abstract
Humanoid loco-manipulation is particularly challenging because it requires coordinating locomotion, balance, and object interaction, making it difficult to learn from limited demonstrations or from scratch through Reinforcement Learning (RL). We introduce DiPLO (self-distilled Diffusion Planning for Loco-manipulation), a framework that combines a self-distilled diffusion planner with a low-level whole-body tracking policy. Our planner is pretrained via supervised learning on motions derived from human-to-humanoid retargeting, while the tracker is trained via RL to execute them. DiPLO centers around two complementary post-training stages. First, we introduce Self-Distillation Diffusion Policy Optimisation (SD:DPO) to remove the planner's dependence on strong task conditioning, which we identify as a crucial ingredient for effective pretraining, as downstream fine-tuning does not recover comparable performance when it is omitted.. We then fine-tune this distilled model with Diffusion-Augmented Proximal Policy Optimization (DA:PPO), a maximum-entropy RL method for diffusion models, to improve task success and align generated plans with the frozen tracker's execution capabilities. Together, these stages enable motion generation through state-history and goal conditioning alone, enabling 99% task success in pick-carry-place tasks, compared to just 45% for the pretrained model. The resulting planner also generates motions for unseen object goals, with further evaluations on kicking and pushing/pulling tasks. Finally, we deploy our pipeline on the hardware, using our DA:PPO planner-tracker on a Unitree G1, achieving repeated successful executions without any real-world fine-tuning.