Neural Sensorimotor control


(a) Chemical signals diffused from the target are sensed by the suckers, based on which the arm needs to catch the target; (b) An abstract block diagram of the sensorimotor control problem.

Neuromuscular arm model

We built models to capture the mechanical properties of arm musculature and the electrical properties of the arm peripheral nervous system (PNS), and the coupling of PNS with muscular contractions. The neural activities of the nerve cord is modeled using cable theory.

Cable-theoretic model of the arm PNS. Cable dynamics take current input and output voltage. Then the muscle activation is generated through the neuromuscular coupling. 

Sensory system - neural rings and sensing units

The arm sensory system is modeled including chemosensing and proprioception. A central component of the sensing neuroanatomy is the model of the neural ring which is used to encode an angle variable.

Neural ring architecture and its properties. The weight function without shift is a Mexican hat function where its positive part create excitatory connections between neurons while the negative part corresponds to inhibition. Without shift, the membrane potential forms a peak. With shift, the peak shifts with speed that depends on the shift of weight function. The peak location is used to encode target bearing and arm local orientation.

Control algorithm - neural motor feedback law

The control law is developed upon the sensory feedback control law we developed earlier in muscle couple level. Now it is extended to neural level as current input to cables. Exact formula will be found in the paper to be published.

Sensing algorithm - consensus

Each sensor has its own estimates of local orientation θi, target bearing αi, and concentration intensity parameter μi for i=1,...,N which create a target estimate ri^target. The goal is for all sensors to reach a consensus on their target beliefs and eventually estimate the true target location (See figure below).

Sensing as a consensus problem

Demonstration Video

Future work

Extend to 3D scenario;

Consider more realistic fluid environment.


Slides from the doctoral preliminary exam of Tixian Wang at University of Illinois  Urbana Champaign, Mar 2024.


T. Wang, U. Halder, E. Gribkova, M. Gazzola, and P.G., Mehta, "Neural Models for Sensorimotor Control of an Octopus Arm," (under review).


Financial support from ONR MURI N00014-19-1-2373, NSF EFRI C3 SoRo #1830881, NSF OAC #2209322, and ONR N00014-22-1-2569.