2019/04/03 by Edward Beeching, Beeching, Edward, Christian Wolf +5
Computer Science · #FOS: Computer and information sciences #Human Pose and Action Recognition #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Multimodal Machine Learning Applications #Reinforcement Learning in Robotics
paper · pdf · doi:10.48550/arxiv.1904.01806
openalex publication_date 2019/04/03 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
An important goal of research in Deep Reinforcement Learning in mobile\nrobotics is to train agents capable of solving complex tasks, which require a\nhigh level of scene understanding and reasoning from an egocentric perspective.\nWhen trained from simulations, optimal environments should satisfy a currently\nunobtainable combination of high-fidelity photographic observations, massive\namounts of different environment configurations and fast simulation speeds. In\nthis paper we argue that research on training agents capable of complex\nreasoning can be simplified by decoupling from the requirement of high fidelity\nphotographic observations. We present a suite of tasks requiring complex\nreasoning and exploration in continuous, partially observable 3D environments.\nThe objective is to provide challenging scenarios and a robust baseline agent\narchitecture that can be trained on mid-range consumer hardware in under 24h.\nOur scenarios combine two key advantages: (i) they are based on a simple but\nhighly efficient 3D environment (ViZDoom) which allows high speed simulation\n(12000fps); (ii) the scenarios provide the user with a range of difficulty\nsettings, in order to identify the limitations of current state of the art\nalgorithms and network architectures. We aim to increase accessibility to the\nfield of Deep-RL by providing baselines for challenging scenarios where new\nideas can be iterated on quickly. We argue that the community should be able to\naddress challenging problems in reasoning of mobile agents without the need for\na large compute infrastructure.\n