The Worldwide Journal of Robotics Analysis, Forward of Print.
Neuromorphic computing mimics computational ideas of the mind in silico and motivates analysis into event-based imaginative and prescient and spiking neural networks (SNNs). Occasion cameras (ECs) solely seize native depth adjustments and provide superior energy consumption, response latencies, and dynamic ranges. SNNs replicate organic neuronal dynamics and have demonstrated potential as options to traditional synthetic neural networks (ANNs), reminiscent of in lowering vitality expenditure and inference time in visible classification. Nonetheless, these novel paradigms stay scarcely explored exterior the area of aerial robots. To research the utility of brain-inspired sensing and information processing, we developed a neuromorphic strategy to impediment avoidance on a camera-equipped manipulator. Our strategy adapts high-level trajectory plans with reactive maneuvers by processing emulated occasion information in a convolutional SNN, decoding neural activations into avoidance motions, and adjusting plans utilizing a dynamic movement primitive. We carried out experiments with a Kinova Gen3 arm performing easy reaching duties that contain obstacles in units of distinct job eventualities and compared to a non-adaptive baseline. Our neuromorphic strategy facilitated dependable avoidance of imminent collisions in simulated and real-world experiments, the place the baseline constantly failed. Trajectory variations had low impacts on security and predictability standards. Among the many notable SNN properties had been the correlation of computations with the magnitude of perceived motions and a robustness to completely different occasion emulation strategies. Exams with a DAVIS346 EC confirmed comparable efficiency, validating our experimental occasion emulation. Our outcomes inspire incorporating SNN studying, using neuromorphic processors, and additional exploring the potential of neuromorphic strategies.
发布者：Ahmed Abdelrahman，转转请注明出处：https://robotalks.cn/a-neuromorphic-approach-to-obstacle-avoidance-in-robot-manipulation/
