Humans often manipulate objects without their fingers alone. We press items against tables, slide them against surfaces, or pivot them using friction from the environment. IRR is exploring a unified control framework that gives Honda’s dexterous robotic hand similar capability: extrinsic dexterity.

Our approach models how forces from both the robot’s fingers and the external environment can be stably estimated from sometimes noisy force readings by utilizing geometric primitives, to accomplish tasks that would otherwise require far more complex hand configurations. This expands the robot’s operational workspace and reduces reliance on intricate joint motions.

By formalizing force estimation framework for intrinsic and extrinsic dexterity, we are making difficult manipulations more robust, more efficient, and more flexible – especially in the cluttered or unstructured environments that future real-world deployment will present.

Learn more in the associated research publication.