Neuromorphic Force Control via Golgi Tendon Organ Models - in collaboration with IIT Genova – Chiara Bartolozzi

alternate text Figure 9 Set up chip an hand and low level control scheme with Spindles and Motorneurons

The current system implements position control using a biologically inspired feedback loop based on muscle spindle afferents. Running on the DYNAP-SE neuromorphic processor, the architecture enables real-time interaction between spiking motor neurons and proprioceptive feedback to regulate the position of a prosthetic finger. This setup demonstrates adaptive control, natural motor unit recruitment, and reflex-like behavior without relying on classical PID control.

To move toward a more complete neuromorphic motor control system, we now propose adding a second level of control based on a spiking model of the Golgi tendon organ (GTO). In biological systems, GTOs monitor muscle force and provide inhibitory feedback to motor neurons, helping regulate tension and prevent damage from excessive force. This mechanism is critical for fine control of grip strength and compliant interaction with objects.

In our proposed extension, a spiking GTO population will be implemented on the same DYNAP-SE chip. It will receive input encoding tendon force—estimated from motor current or embedded force sensors—and project to inhibitory interneurons that modulate the motor neuron pool. This feedback will mimic autogenic inhibition, allowing force regulation to emerge naturally from the neuromorphic circuitry.

By integrating both position and force feedback in a fully event-driven architecture, the system will support more robust, adaptive, and biologically plausible prosthetic control. The dual-loop setup will enable compliant and context-aware motor output, paving the way for multi-joint prosthetic hands capable of interacting more naturally with users and the environment. This work is conducted in collaboration with the IIT Genova team, leveraging their expertise in neuromorphic sensing and embodied soft robotics.

 

Methodology

  • - Implement a spiking GTO model on the DYNAP-SE neuromorphic chip, emulating force-sensitive inhibitory afferents.

  • - Estimate finger force via motor current sensing or embedded soft force sensors, and encode this information into spike trains.

  • - Route GTO output to inhibitory interneurons connected to alpha-motor neurons, implementing biologically inspired autogenic inhibition.

  • - Integrate the new force-feedback loop with the existing spindle-based position control architecture for simultaneous modulation.

  • - Validate the dual-loop control system on a single finger of the Mia prosthetic hand during varying load and grip conditions.

 

Expected Outcomes

  • - Acquired hands-on expertise with neuromorphic control using mixed-signal chips and robotic actuation.

  • - Functional implementation of a GTO-based feedback loop for force regulation.

  • - Demonstrated improvements in compliance, adaptability, and safety during prosthetic finger control.

 

Available material

  • - Chip tutorial and literature

  • - Hand control coding

 

Requirements

Python programming for working with Dynapse, familiarity with signal processing and neural networks

 

Contacts

Elisa Donati (elisa@ini.uzh.ch)

 

References

Dideriksen J L, Negro F, Farina D (2015) The optimal neural strategy for a stable motor task requires a compromise between level of muscle cocontraction and synaptic gain of afferent feedback. Journal of neurophysiology, 114(3), 1895-1911.

Niu, Chuanxin M., Sirish K. Nandyala, and Terence D. Sanger. "Emulated muscle spindle and spiking afferents validates VLSI neuromorphic hardware as a testbed for sensorimotor function and disease." Frontiers in computational neuroscience 8 (2014): 141.

Mileusnic, Milana P., and Gerald E. Loeb. "Mathematical models of proprioceptors. II. Structure and function of the Golgi tendon organ." Journal of neurophysiology 96.4 (2006): 1789-1802.

back
back