REINFORCEMENT LEARNING APPLICABILITY FOR TRAINING OF THE DISTRIBUTED COMPUTER VISION MODELS IN UAVS

Fairly, the distributed computer vision models (DCV) used in UAVs are still a new scientific area, but they already have some directions for research.The first direction covers techniques of accelerating reinforcement learning RL through parallel and distributed computing, including simulator parallelism, computational parallelism, synchronisation, libraries, etc.It is critically important for solving visual tasks, reducing wall-clock time wasting and stabilising optimisation policies where larger parent experiences and large networks are required.[1] In decentralised or multi-agent scenarios, RL "stitches" vision and action across multiple nodes, solving joint scene perception and coordination with partial observability.[2] The second direction covers edge learning and emphasises that it is distributed approaches that provide the ability to train models based on video streams from a camera directly following a secure noncloud route.[3]