Reading focus
Syllabus: How Do Neural Circuits Turn Sensation into Action?
Follow how prediction, state estimation, and feedback address uncertainty during movement.
Why you’re reading it. This asks what a motor circuit must compute, extending the physiological account without equating a model diagram with anatomy.
A question to consider. For the stumbling example, what observation would distinguish inaccurate state estimation from faulty coordination of otherwise appropriate commands?
Why this reading
To control a movement, an organism must act despite delayed, noisy information and changing physical conditions. Wolpert and Ghahramani organize these problems around prediction, estimation, and control. Their framework makes explicit what a proposed motor mechanism would need to accomplish.
Before you read
A forward model predicts consequences from a state and a command; an inverse model works from a desired consequence toward a command. An efference copy is information about an outgoing motor command available to other processing. Feedback uses observed consequences to adjust action. A state estimate combines imperfect information about the body and environment. An optimal solution is optimal relative to a specified objective and its assumptions, not automatically best in every sense.
Read for
- Distinguish estimating the current state from selecting an action and learning the properties of a new context.
- Follow how predicted and observed sensory consequences can be combined when feedback is delayed or uncertain.
- Ask what the boxes and arrows in the models specify computationally, and what additional evidence would identify a neural implementation.
Consider afterwards
- You lift a carton that is much lighter than expected. How would prediction, sensory feedback, and context estimation contribute at different moments?
- Two models predict the same reaching trajectory. What perturbation or neural measurement might distinguish how they achieve it?
Editorial context
This review develops a particular computational approach through selected examples. Its model diagrams are hypotheses about operations, not anatomical maps, and a good behavioral fit alone does not establish their physiological implementation. Distinguishing an objective, a representation, and an implementation makes these proposals testable without assuming that every movement must solve the same optimization problem.