Search papers, labs, and topics across Lattice.
This paper introduces DreamControl-v2, an improved framework for training autonomous humanoid skills that trains a guided diffusion model directly in the robot's motion space, unifying human and robot datasets. By training a diffusion prior in the robot's embodiment space, the method captures a wider range of skills and automates the pipeline by removing manual filtering. Experiments in simulation and on a real Unitree-G1 robot validate that scaling the generation of reference trajectories is crucial for robust downstream RL policies.
Training diffusion priors directly in a robot's motion space, using a mixture of human and robot data, unlocks more diverse and robust humanoid skills without manual intervention.
Developing robust autonomous loco-manipulation skills for humanoids remains an open problem in robotics. While RL has been applied successfully to legged locomotion, applying it to complex, interaction-rich manipulation tasks is harder given long-horizon planning challenges for manipulation. A recent approach along these lines is DreamControl, which addresses these issues by leveraging off-the-shelf human motion diffusion models as a generative prior to guide RL policies during training. In this paper, we investigate the impact of DreamControl's motion prior and propose an improved framework that trains a guided diffusion model directly in the humanoid robot's motion space, aggregating diverse human and robot datasets into a unified embodiment space. We demonstrate that our approach captures a wider range of skills due to the larger training data mixture and establishes a more automated pipeline by removing the need for manual filtering interventions. Furthermore, we show that scaling the generation of reference trajectories is important for achieving robust downstream RL policies. We validate our approach through extensive experiments in simulation and on a real Unitree-G1.
