Real-Time Control of a Humanoid Robot for Whole-Body Tactile Interaction

Enabling robots to interact physically with complex, unstructured environments remains a significant challenge. Methods relying solely on joint-torque sensing suffer from ambiguity in multicontact scenarios, while vision is prone to occlusion. This article presents an approach using a whole-body tactile skin sensor network to address these limitations by integrating feedback for compliance, force control, and collision avoidance. The control framework uses quadratic programming to integrate rich tactile and proximity feedback from the skin network. To maintain real-time performance with this dense sensory data, the method clusters sensor activations into active regions. This enables the robot to generate whole-body compliance, regulate interaction forces across various body parts, and transform proximity feedback into distance constraints to dynamically avoid collisions in unmodeled environments. This approach's effectiveness and real-time feasibility through experiments on a humanoid robot performing tasks such as transporting bulky objects, directly controlling interaction forces, and avoiding dynamic obstacles are demonstrated.