Abstract:

It is common to equate robot safety with “collision avoidance”, but in unstructured open-world environments, a robot’s representation of safety should be much more nuanced. For example, the household manipulator should understand that pouring coffee too fast will cause the liquid to overflow or pulling a mug too quickly from a cupboard will cause other dishes to fall and break. While in theory, traditional methods from control theory can synthesize safe policies for general nonlinear systems and nonconvex constraints, in practice, it has been limited to hand-engineered collision-avoidance constraints modeled via low-dimensional state-space representations and first-principles dynamics. In this work, our goal is to generalize safe robot controllers to prevent failures that are hard — if not impossible — to write down by hand, but can be intuitively identified from high-dimensional observations: for example, spilling the contents of a bag. We propose Latent Safety Filters, a latent-space generalization of Hamilton-Jacobi reachability that tractably operates directly on raw observation data (e.g., RGB images) by performing safety analysis in the latent embedding space of a generative world model. This transforms nuanced constraint specification to a classification problem in latent space and enables reasoning about dynamical consequences that are hard to simulate. In simulation and hardware experiments, we use Latent Safety Filters to safeguard arbitrary policies (from generative policies to direct teleoperation) from complex safety hazards, like preventing a Franka Research 3 manipulator from spilling the contents of a bag or toppling cluttered objects.

Committee:
Andrea Bajcsy (advisor)
Max Simchowitz
Zac Manchester
Gokul Swamy (PhD Student)