Learning for Microrobot Exploration: Model-based Locomotion,\n Sparse-robust Navigation, and Low-power Deep Classification

2020/04/27 by Nathan Lambert, Farhan Toddywala, Lambert, Nathan O. +9
Engineering · Physics and Astronomy · #FOS: Computer and information sciences #Micro and Nano Robotics #Modular Robots and Swarm Intelligence #Robotics (cs.RO) #Robotics and Sensor-Based Localization

paper · pdf · doi:10.48550/arxiv.2004.13194

openalex publication_date 2020/04/27 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Building intelligent autonomous systems at any scale is challenging. The\nsensing and computation constraints of a microrobot platform make the problems\nharder. We present improvements to learning-based methods for on-board learning\nof locomotion, classification, and navigation of microrobots. We show how\nsimulated locomotion can be achieved with model-based reinforcement learning\nvia on-board sensor data distilled into control. Next, we introduce a sparse,\nlinear detector and a Dynamic Thresholding method to FAST Visual Odometry for\nimproved navigation in the noisy regime of mm scale imagery. We end with a new\nimage classifier capable of classification with fewer than one million\nmultiply-and-accumulate (MAC) operations by combining fast downsampling,\nefficient layer structures and hard activation functions. These are promising\nsteps toward using state-of-the-art algorithms in the power-limited world of\nedge-intelligence and microrobots.\n

Related

• Robot Navigation using Reinforcement Learning and Slow Feature Analysis
• Tiny Robot Learning: Challenges and Directions for Machine Learning in Resource-Constrained Robots
• Micro/Nano Motor Navigation and Localization via Deep Reinforcement Learning
• From Handheld to Unconstrained Object Detection: a Weakly-supervised On-line Learning Approach
• Towards Cognitive Exploration through Deep Reinforcement Learning for Mobile Robots