Self-Organizing Sensory-Motor Map for Low-Level Touch Reactions

After being turned on, the robot does not know how to react on tactile stimulation on its artificial sensor skin units.
Our aim is to automatically find an appropriate reflex like reaction.
Our algorithm therefor only receives knowledge on the number of sensor units it can read from and degrees of freedom it can actuate.
Here we distributed 15 multi-modal sensor units (HEX-O-SKIN) on a KUKA light weight robotic arm with 7 degrees of freedom.

The robot then explores the current pose through sinusoidal velocity pattern, on one degree of freedom after the other.
During this exploration, data from the accelerometer on every unit is being sampled.
At the end of the exploration, our algorithm assembles a new part of the sensory-motor map, related to the explored pose.
This part is then stored in memory and can later on be used to map tactile stimulation in the same or similar poses.
So far we only focused on lateral weights and lateral stimulation -- this means in the direction of the surface normal.

Our approach is to then use the sensory motor map in the opposite direction, transferring touch events detected on the sensor unit level to motor commands.
Thus the robot knows how to react on contact, detected by four GP2S60 proximity sensors on every sensor unit.
The behavior is defined for a single sensor unit and instantiated as often as there are sensor units distributed on the robot.
Multiple reactions on a unit are super posed, as well as the degrees of freedom velocity commands generated on mapping all sensor unit reactions.

P. Mittendorfer and G. Cheng, "Self-organizing sensory-motor map for low-level touch reactions," 11th IEEE-RAS International Conference on Humanoid Robots, (accepted), October 2011.

P. Mittendorfer and G. Cheng, "Humanoid multi-modal tactile sensing modules," IEEE Transactions on Robotics, vol. 27, no. 3, pp. 401--410, June 2011.

Category:

Science & Technology

Tags:

• Sensory-Motor Map
• Touch Reactions
• Artificial Skin

[mezcestro]

I am glad the video shows how the robot learns.

The machine acts as if alive! 0:48