Neural Gait Synthesis for Autonomous Biped Robots

该奖项资助的方法集成的初步研究， 反射性步态合成和对困难地形的适应， 一个能自由行走的机器人 这项研究将探索神经- 用于训练中央模式生成器的基于网络的机制 生成一个新的步态， 学习和适应单元。 它还将研究收敛 反向传播和强化学习的性质 学习和适应股使用的机制。 的 这项研究的长期目标是实现 比多足行走更灵活机动 机器或其他形式的机器运动，并最终 为移动的机器人提供增强或替代人类的能力 在危险的操作或崎岖的地形。

This award funds initial research on integration of methods for reflexive gait synthesis and adaptation to difficult terrains, for a biped walking robot. The research will explore neural- network-based mechanisms for training a central pattern generator to generate a new gait that is learned by a terrain-monitoring learning and adaptation unit. It will also study the convergence properties of backpropagation and reinforcement learning mechanisms for use in the learning and adaptation unit. The long-term goal of this research in biped robotics is to achieve more agility and mobility than is possible in multilegged walking machines or other forms of machine locomotion, and ultimately to provide capability for mobile robots to augment or replace humans in hazardous operations or in rough terrain.