Extended Data Fig. 1: Autonomous bionic technology and control design.
From: Continuous neural control of a bionic limb restores biomimetic gait after amputation
a, A photograph of the autonomous neuroprosthetic system. b, The neuroprosthetic interface for bionic ankle control consists of the Agonist-antagonist Myoneural Interface (AMI) comprising flexible electrodes as the sensing modality for the AMI implementation. c, Portable EMG sensor unit. d, Autonomous powered prosthetic ankle. e, Control design enabling direct neuromodulation of bionic ankle movements. The controller processed the dorsiflexor and plantar flexor electromyography (EMG) signals to compute the EMG envelope. Using the EMG envelope, a decoding pipeline continuously decoded motor intention into target equilibrium joint angle (\({\theta }_{{\rm{ref}}}\)) and joint impedance modulation level (\({\mu }_{Z}\)). The ankle actuation direction was obtained through the current joint angle (\(\theta\)) and \({\theta }_{{\rm{ref}}}\), which was used to account for differences in dorsiflexion (DF)-plantar flexion (PF) motor coordination during the \({\mu }_{Z}\) normalization process. The joint torque command was determined based on an impedance control scheme78,79 which modeled joint torque using joint impedance (\({K}_{\theta }\), \({K}_{v}\), \({K}_{{\rm{psv}}}\), \({D}_{{\rm{psv}}}\)), a neurally-controlled impedance scaling factor \({\mu }_{Z}\), and a neurally controlled equilibrium \({\theta }_{{\rm{ref}}}\). We designed the joint impedance to emulate the biological ankle angle- and velocity-dependent upper torque bound for any measured values of \(\theta\) and \(\dot{\theta }\). Note that the controller did not involve any intrinsic gait control features like state machines or pattern recognition algorithms for enabling biomimetic gait.
