Multimode Bio-Inspired Sense and Avoid
Abstract
The Royal Veterinary College (RVC) and Centeye have, in parallel, developed devices that offer capability enhancement in the sphere of sense-and-avoid technology for micro- and nano-rotorcraft. Specifically, the RVC devices use information measured from changes in the induced flow field when the craft enters ground or wall effect, while Centeyes devices use visual information, including optical flow, stereo vision, and active illumination. Each sensory mode has been studied, flight tested, and characterized extensively in recent efforts. Here, we exploit these capabilities more deeply by feeding the sensor array measurements into a deep learning network to estimate bearing and distance to nearby obstacles in a range of conditions, including those that are challenging for one or the other sensory mode. Avoidance behaviour will be superior to that which is possible by each mode independently. Our objectives included integration of these two technologies onto a suitable test platform for flight testing, simultaneous acquisition of data from at least two sensor modes, and first-order attempts at algorithm fusion in supervisory flight control software. The integrated sensory mode platform demonstrates obstacle avoidance greater than the sum of its parts. Additionally, we developed attention-switching algorithms to reduce power consumption under differing environmental conditions, and the identification of new capabilities other than obstacle avoidance using the same apparatus, such as the measurement of a wind vector. Our machine learning method matches the physical arrangement of the sensors but is fundamentally agnostic to the sensor type and is applicable to acoustic range finding using inherent vehicle noise as the signal.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 11, 2023
- Accession Number
- AD1224976
Entities
People
- Richard J. Bomphrey
Organizations
- Royal Veterinary College, University of London