insideOut: A Bio-Inspired Machine Learning Approach to Estimating Posture in Robots Driven by Compliant Tendons

Abstract

Estimates of limb posture are critical for controlling robotic systems. This is generally accomplished with angle sensors at individual joints that simplify control but can complicate mechanical design and robustness. Limb posture should be derivable from each joint's actuator shaft angle but this is problematic for compliant tendon-driven systems where (i) motors are not placed at the joints and (ii) nonlinear tendon stiffness decouples the relationship between motor and joint angles. Here we propose a novel machine learning algorithm to accurately estimate joint posture during dynamic tasks by limited training of an artificial neural network (ANN) receiving motor angles and tendon tensions, analogous to biological muscle and tendon mechanoreceptors. Simulating an inverted pendulum—antagonistically-driven by motors and nonlinearly-elastic tendons—we compare how accurately ANNs estimate joint angles when trained with different sets of non-collocated sensory information generated via random motor-babbling. Cross-validating with new movements, we find that ANNs trained with motor angles and tendon tension data predict joint angles more accurately than ANNs trained without tendon tension. Furthermore, these results are robust to changes in network/mechanical hyper-parameters. We conclude that regardless of the tendon properties, actuator behavior, or movement demands, tendon tension information invariably improves joint angle estimates from non-collocated sensory signals.

Edge Computing in Nature: Minimal pre-processing of multi-muscle ensembles of spindle signals improves discriminability of limb movements

“…Recent computational work suggest that spindle information is likely necessary to extract a percept of body configuration (Berry et al, 2017;Berry and Valero-Cuevas, 2020;Sandbrink et al, 2020), but not sufficient as it may need to be conditioned on tendon tension information from Golgi tendon organs (Hagen et al, 2021). Notwithstanding the geometrically obligatory relationship between joint angles and musculotendon lengths, the tenet that spindle information is a primary source of body configuration (even when conditioned on Golgi tendon organ signals) remains to be proven as musculotendons often span multiple joints (Valero-Cuevas, 2016) and spindle signals can be modulated independently of joint angles by γ− motoneuron drive to their intrafusal fibers in which the mechanoreceptors sit (Jalaleddini et al (2017) and references therein).…”

Developing Intelligent Robots that Grasp Affordance

“…Fetal-like random muscle twitches resulted in stable connectivity patterns ( Figure 6D ) that were similar to classical descriptions of proprioceptive spinal circuitry, including the homonymous monosynaptic stretch reflex, reciprocal length feedback, inhibitory force feedback and convergence of force and length feedback in excitatory premotor interneurons ( Pierrot-Deseilligny and Burke, 2005 ; Alstermark et al, 2007 ). Similar motor babbling has been applied successfully to account for learned perception of body posture from physiological models of proprioceptors that project to the thalamocortical brain ( Hagen et al, 2021 ). Cutaneous circuits are much less well characterized in animals but are known to be essential for dexterous manipulation, both to modulate grip forces and to trigger different phases of tasks ( Johansson and Flanagan, 2009 ).…”