Slow Computing Simulation of Bio-plausible Control
Abstract
In order to implement control methods on low power, size-constrained systems, significant departures from standard approaches are needed. We look to biology for inspiration and focus on the problem of attitude stabilization through visual input alone. Bio-plausibility broadly refers to the ability of an operation to be carried out on a neural substrate or a parallel asynchronously operating computation scheme. One method of realizing such an asynchronous, parallel computing capability in power-constrained systems is by making use of low-throughput, energy-efficient computing elements; this is referred to as a "slow computing" architecture. In this work, we evaluate the capacity of bio-plausible control laws for attitude stabilization on a slow computing architecture. The Gillespie Stochastic Simulation Algorithm (SSA) from chemical kinetics is modified to simulate a slow computing system. We approximate how sparse the system's visual information may be while remaining controllable and producing attitude stabilization. Gaussian noise is added to the simulation of visual input values, and the interaction of noise and sparsity is explored. The bio-plausible approach to stabilization is intended to replace more power-consumptive methods. In simulation, low power processing of visual input data produced effective attitude stabilization for a range of noise and sparsity.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 2012
- Accession Number
- ADA559285
Entities
People
- Alec Koppel
- Alma Wickenden
- Brian M. Sadler
- Robert Proie
- Vishnu Ganesan
- William Nothwang
Organizations
- United States Army Research Laboratory