Sensorimotor Integration: Perception-action relationship

Question 1

What is sensorimotor integration? Can you give an example for a model for sensorimotor integration?

Answer 1

Sensorimotor integration is a process in the brain that produces task-specific motor output based on selective and rapid integration of sensory information from multiple sources.
Output is motor commands

Simple model for sensorimotor integration: Braitenberg Vehicles

Question 2

What are Braitenberg vehicles? Can you give an example?

Answer 2

Simple reactive control: The sensors are directly connected to motor commands (actuators)

Connection types:
- Ipsilateral (sensors are connected to actuators on the same side)
- Contralateral (sensors are connected to actuators on opposite side)

• Excitatory (sensor reading & signal to actuator are proportional): Sensor value goes up => motor value goes up.
• Inhibitory (sensor reading & signal to actuator are inverse proportional): Sensor value goes up => motor value goes down.

Question 3

What are the drawbacks of Braitenberg vehicles? How can we improve Braitenberg control?

Answer 3

Drawbacks:
○ Noise highly affects the output (since input is directly linked to the output)
○ Reactive => no predicting or learning

Improve Braitenberg control: Use forward models to predict and correct sensor values.

Question 4

What are sensorimotor contingencies (SMCs)? How can we exploit SMCs?

Answer 4

• Relationship between movements and observed changes in sensory input.
• SMCs capture regularity (i.e., stable and predictable) in how sensory information depends on movement (motor commands) and vice versa.

Example: Foot ball coming towards you:
○ Estimate where the ball might hit (chest)
○ Anticipation - has idea of the force of the impact from the ball hitting the chest.
○ Predict how much the force sensor will impact the balance and and make anticipatory action (move based on this prediction) to compensate for the impact (before the ball hits the chest) in order to not loose balance (i.e. proactive).
- Sensory part: Predictable displacement of ball, vision, impact, loss of balance
Motor part: Anticipatory action (compensate for anticipated impact force form ball)

Exploit SMCs:
Use SMC’s to predict the outcome of executing the plan from the inverse model (e.g., motor command) at each processing step
- Error between predicted outcome and actual outcome is used to is change response of the inverse sensory model = improve the model
=> anticipatory system that compensate/deals with delays in feedback system.

Question 5

What are internal models? Can you give an example?

Answer 5

Inter-dependence between sensory information and motor commands is captured by internal models

The Internal model of a system simulates the response of the system to its input.

Example: Hit a table with a hammer

Question 6

What is a forward/inverse model? Can you give an example?

Answer 6

• Inverse Model: Input desired goal, output a plan to achive the goal (anticipatory action). They are limited by feedback delay.
• Forward Model: Takes a copy of IMs plan as input and predicts what happens when executed. Gives fast response if feedback is delayed

Example: Goal: Hit a table with a hammer. Inverse model makes a plan to achieve the goal and sends it to the actuators. Forward model takes a copy of the plan and predicts the sensor values after the plan has been executed.

Question 7

What are the useful characteristics of forward models?

Answer 7

• Feedback from forward model allows one to predict and use sensory outcome of an action before external sensory feedback from the environment becomes available (i.e., state estimation) => Compensate for alere delays in control loop.
• Correct for errors by transforming them into motor commands
• Predict and cancel sensory effects of moving
• Predict input to inverse model but also change the inverse model itself (learning) => adaptive control