Human-like motion is characterized by a straight or slightly curved trajectory with a smooth, bell-shaped end-point velocity. In order to achieve such a human-like point-to-point reaching motion with a robot, the best approach would be to create an actuation system that mimics key aspects of the human actuation system, specifically the nervous and musculoskeletal systems. The actuator type selected for the robot must be compliant and form force-generating subunits, much like human skeletal muscle. A key feature of human motion is the presence of motor variability and consequent errors in task performance. This suggests some form of randomness in the neuromuscular system. Harris and Wolpert first presented the concept of signal dependent noise (SDN), in which motor variability is caused by noise that is linearly proportional to the magnitude of the motor command. This can be modeled as multiplicative noise in the system. A straightforward implementation of SDN in a robot would require the introduction of a Gaussian noise generator; however, this is not preferable from a hardware perspective because it requires additional componentry. Further, no such extraneous noise generator has been found in the biological system. Therefore, an alternative, biologically inspired method of introducing SDN into the system was needed. Floating point quantization (FPQ), a popular numbering scheme in digital communications, appeared as a viable option which mimics aspects of Henneman’s size principle of motor unit recruitment. This paper presents a non-uniform recruitment method for compliant actuator arrays. FPQ results in larger deviations from the original signal as amplitude increases, in the same manner as SDN.

This content is only available via PDF.
You do not currently have access to this content.