Theory 11/Motor Control & Movement /Motor neuroscience/3 min
The Brain Stores Stiffness, Not Paths
We talk about motor programmes as if the brain recorded a path. It may be recording a spring.
Motor memory is stored as an impedance profile rather than a trajectory, so what is learned is how rigidly to resist perturbation rather than where to move, and transfer between tasks should follow stiffness similarity rather than kinematic similarity.
01The evidence
- Bioinspired smooth neuromorphic control for robotic arms implements movement through impedance rather than explicit trajectory, which demonstrates that smooth behaviour can be generated from a stiffness policy.Bioinspired neuromorphic control
- Minimum-jerk formulations combined with feedback control show that smooth trajectories can emerge from a controller that regulates impedance, without storing the path explicitly.IET Control Theory
02The tension
Motor learning is described in terms of trajectories, and transfer is predicted from kinematic similarity: practice a movement and improve a similar one. Transfer in practice is patchy, and it follows the dynamics of the task more than the shape of the path.
03The theory
The nervous system controls movement by setting the mechanical properties of the limb, notably stiffness at each joint, and letting the environment and the reflex loops produce the trajectory. What is learned is the stiffness profile, not the path.
That makes transfer a function of dynamic similarity. A task that requires the same impedance will transfer, even if the path is different, and a task with the same path but different impedance demands will not. This is a different prediction from the kinematic account and it is testable with force fields.
The theory also explains why smoothness is so robust. A stiffness policy with feedback produces smooth trajectories automatically, so smoothness is a consequence of the control scheme rather than an objective that is optimised.
The applied consequence is that rehabilitation should train impedance, for instance with perturbation and load, rather than rehearse trajectories, and that transfer should be predicted from the dynamics of daily tasks.
04What this is not
Those accounts store a path and predict transfer from kinematic similarity. This stores an impedance and predicts transfer from dynamic similarity, which is why path-similar training often fails to transfer.
05Predictions
- P1Transfer should follow stiffness similarity better than kinematic similarity.
- P2Training with force fields should transfer to tasks with different paths but similar impedance demands.
- P3Smoothness should emerge without an explicit smoothness objective.
06What would refute it
Transfer tracking path similarity at matched dynamics would restore the trajectory account.
07What it would change
It would change rehabilitation design and how motor skill is assessed.
08Evidence base
- Bioinspired smooth neuromorphic control for robotic arms — arXivhttp://xxx.itp.ac.cn/abs/2209.02787v1
- Minimum jerk norm with feedback control — IET Control Theoryhttps://ietresearch.onlinelibrary.wiley.com/doi/pdf/10.1049/iet-cta.2016.0220