Path planning using a neuron array integrated circuit

Abstract

Neuromorphic Engineering is an interdisciplinary field which combines concepts from fields such as biology, neuroscience, computer science and engineering. The goal of this field is to design systems that are based on the principles of biological nervous systems. This paper presents hardware results for path planning using a neuron array integrated circuit. The algorithm is explained and experimental results are presented showing 100% correct performance for a large number of maze environment scenarios. Although there is still more work to be completed before this is a fielded system, this work represents an new application of a neuromorphic Integrated Circuit and the results demonstrate definite potential.Index Terms-aVLSI, path planning, robotics, neuromorphic 1. ANA LOG VLSI NEURON EMBEDDED SYSTEM FOR PA TH PLANNING Path planning is a critical task for robots, autonomous vehicles, animated characters, etc. Fig 1 is a cartoon showing the ultimate goal of the problem being addressed in this paper, namely how to use a neuron array integrated circuit (IC) [2] to plan a path for a Micro Aerial Vehicle (MAV) (or similar power constrained ground or sea robot) through an environment in an effort to conserve its limited battery resources. The IC used for the results in this work, Fig 2a, uses biologically realistic transistor based models which operate how neurons do, however not necessarily how the brain does path planning. The IC has 100 Neurons and 30,000 synapses. It was constructed in a 0.35 micron process and the die size is 5x5 mm. Floating gate transistors are used as the synapse elements. The floating gate synapse transistors are used to create the programmable routing. A synaptic weight is stored by an adjustable charge on the gate of the floating gate synapse transistor. Multiple synapses are connected to a dendrite by adjusting the weights on the synapse transistors which are connected in parallel to the dendrite wires. Unused synapses are not connected internally to the path planning circuit. The floating gates for these transistors are not set to conduct. The system uses Address Event Representation (AER) to record spike data from the neurons. Neuron Elements include: Soma, dendrite, synapses, and axons. Path planning can be summarized with the following three tasks given that states, actions, an initial state, and a goal state are provided. The robot should:1) Find a sequence of actions that take the robot from its Initial state to its Goal state Robot Start a) Discretized locations in maze A -. C -. I I T T b) Grid representation of maze c) Neuron representation of maze Fig. 1. The goal of this research: To use a reconfigurable neuron Array IC to plan a path for small robot from point A to point Z through an environment. a) Maze environment which is discretized into grid points b) Simplified grid representation of the maze in (a). Hash marks denote in-active edges (i.e. walls)c) Bidirectionally connected neurons implementing the edges between nodes.2) Find actions that take the robot from any state to t...