Verifiable Learned Behaviors via Motion Primitive Composition: Applications to Scooping of Granular Media
This work addresses the need for flexible and verifiable robotic systems in industrial applications, though it is incremental as it builds on existing motion primitive and verification methods.
The paper tackles the problem of generating reliable robotic behaviors from natural language inputs by developing a framework where learned behaviors are verifiable by construction, using motion primitive composition and probabilistic verification. They demonstrated this in simulation and on hardware with a robot scooping granular media, achieving real-time behavior generation.
A robotic behavior model that can reliably generate behaviors from natural language inputs in real time would substantially expedite the adoption of industrial robots due to enhanced system flexibility. To facilitate these efforts, we construct a framework in which learned behaviors, created by a natural language abstractor, are verifiable by construction. Leveraging recent advancements in motion primitives and probabilistic verification, we construct a natural-language behavior abstractor that generates behaviors by synthesizing a directed graph over the provided motion primitives. If these component motion primitives are constructed according to the criteria we specify, the resulting behaviors are probabilistically verifiable. We demonstrate this verifiable behavior generation capacity in both simulation on an exploration task and on hardware with a robot scooping granular media.