Neuro-Autonomy: Neuroscience-Inspired Perception, Navigation, and Spatial Awareness for Autonomous Robots
Abstract
The proposed instrumentation will support research by an international team of neuroscientists,computer scientists, and engineers w ho are working to better understand the neurobiology of animalmotion control. While the equipment will be deployed at Boston Univer sity, it is expected to playa significant role in enhancing research being conducted at multiple universities in the U.S. andAustr alia under a MURI grant (N000141912571) whose focus is on neuroscience-inspiredperception, navigation and spatial awareness for autonomous robots. The proposed equipmentwill support three distinct thrusts in the research: (i) it will provide computational re sources foranalyzing large sets of data recorded from moving laboratory animals and humans performingnavigation tasks, (ii) it wil l support video imaging on laboratory animals that will be synchronizedwith neural recordings, and (iii) it will provide a testbed of state-of-the-art vision-enabled mobilerobots that will use neural models for navigation in an existing 1,500 square foot experi mental arena.Specifically:(i) We proposed to enhance our computational resources by acquiring a node on the MassachusettsGreen Hi gh-Performance Computing Center (MGHPCC)with which the researchteam has prior experience. The MGHPCC is a research-focused data ce nter that provides highperformancecomputing functionality to major New England research universities including BostonUniversity an d MIT (both institutions being participants in the neuroautonomy MURI). A nodeon the MGHPCC will provide the peta-byte scale data s torage and computing power that will beneeded as the team works to synchronize high-resolution thermal videos of running animals wi thneurophysiological recordings of the animals brain activity. This MGHPCC node will be usedfor batch processing and will be supp lemented by stand-alone Lambda Deep Learning Workstationswith GPU support to provide similar computing power but with the possibili ty of being usedinteractively. The workstations will be used with state-of-the-art machine learning software thatincludes TensorFl ow, Keras, and PyTorch. They will also be used for real-time interpretation ofcomputationally intensive video streams from mobile r obots described in (iii) below.(ii) The research will leverage experience gained in earlier work in recording and analyzingthermal videos of moving animals. The goal in the current project is to create records of limbmovements and especially head directions tha t can be synchronized and compared with neurophysiologicalrecordings of laboratory animals running through various tailored experim ental settings.The proposed camera equipment is a current version of older equipment with which the team hasexperience and which i ncludes essential current generation software as well as interoperability withother laboratory instrumentation.(iii) A major goal of the teams neuroautonomy research is translation of recorded neurophysiologicaldata from animals to sensory-motor models for mo bile robot control that recreate animalmovement and behaviors in a way that is faithful to the underlying neurobiology. The goal is ey technologies are LIDAR (light detection and ranging), stereocameras (for depth perception), and high ate standard light camerasmainly for optical flow sensing. Important components of the proposed robot systems are onboardcomputer s capable of providing real-time execution of computationally demanding computer visionsoftware. The research will explore sensory- motor control based on spiking neuron models witha long-range goal of transitioning computations to neuromorphic computer hardware, such as theIntel Loihi platform.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 03, 2021
- Source ID
- N000142112844
Entities
People
- Roberto Tron
Organizations
- Boston University
- Office of Naval Research
- United States Navy