A predictive coding framework for safe and versatile control of supernumerary robotic limbs

Abstract

Supernumerary robotic limbs (SLs) have the potential to extend human sensorimotor capabilities by increasing effective degrees of freedom. However, current systems are designed for a single task and cannot flexibly coordinate with natural limbs (NLs). Here we propose a hierarchical control architecture for human–SL augmentation that draws on predictive coding to couple intent inference, safety, action planning, and interaction. The top layer infers user goals and imposes task- and environment-level constraints to guarantee dynamic safety and multilimb coordination. A mid layer plans actions and synthesizes multisensory feedback for the user, while a low layer executes compliant interaction with impedance control and augmented sensory cues. Central to the framework is General Voluntary Control (GVC), a novel interaction mechanism that integrates autonomous SL behaviors with direct human commands. GVC treats the human and SLs as agents engaged in a differential game, sharing motion plans and allocating effort according to (i) a parameter capturing trust in the human plan, and (ii) an effort-sharing parameter that distributes task load. The GVC is validated through simulations of human-SLs object co-manipulation. Across a wide range of movement durations, loads, prediction errors, and diverging motion plans, the controller remains stable, shares effort appropriately, and maintains low tracking error. Human-like impedance adaptation further reduces competition and improves convergence. The results provide design principles for safe, versatile, and intuitive SL control in industrial, assistive, and surgical settings.