It's an exciting time to be doing research in generalized autonomy for humanoid robotics. The following are some of the open problems CRΔL is focusing on:

1. Human-Robot-Environment Co-optimization (HERO). While necessary as a key component, physical AI alone is insufficient for autonomous humanoid robots, particularly when the environment is unstructured, messy, and complex. Going forwards, we will need to co-optimize the robot’s environment in parallel with its autonomy.
2. Generalizable autonomy in open worlds. We're looking at ways to minimize the sim2real gap via real2sim2real. Specifcally, we're designing on-the-fly world models from RGB videos that can be used for post-training off-the-shelf policies.
3. Agile whole body feedback-control. Wheeled robots are easier to control since the action space is two-dimensional. Humanoid robots are not only high-dimensional, but also underactuated. Therefore, in addition to motion planning, one must also solve whole body control problems with feedback.
4. Multi-robot collaboration with humanoid robots. Can multiple humanoid robots collaborate autonomously on difficult tasks? Can they coordinate and resolve conflicts? Do the benefits of expanded mobility capabilities of a single humanoid exponentially multiply when multiple of them work together?