Deep Reinforcement Learning for Single-Shot Diagnosis and Adaptation in\n Damaged Robots

2019/10/02 by Shresth Verma, Haritha Nair, Verma, Shresth +7
Computer Science · #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Reinforcement Learning in Robotics #Robotics (cs.RO)

paper · pdf · doi:10.48550/arxiv.1910.01240

openalex publication_date 2019/10/02 · openalex created_date 2022/07/28 · openalex updated_date 2026/07/28

Abstract

Robotics has proved to be an indispensable tool in many industrial as well as\nsocial applications, such as warehouse automation, manufacturing, disaster\nrobotics, etc. In most of these scenarios, damage to the agent while\naccomplishing mission-critical tasks can result in failure. To enable robotic\nadaptation in such situations, the agent needs to adopt policies which are\nrobust to a diverse set of damages and must do so with minimum computational\ncomplexity. We thus propose a damage aware control architecture which diagnoses\nthe damage prior to gait selection while also incorporating domain\nrandomization in the damage space for learning a robust policy. To implement\ndamage awareness, we have used a Long Short Term Memory based supervised\nlearning network which diagnoses the damage and predicts the type of damage.\nThe main novelty of this approach is that only a single policy is trained to\nadapt against a wide variety of damages and the diagnosis is done in a single\ntrial at the time of damage.\n

Related

• Fault-Aware Robust Control via Adversarial Reinforcement Learning
• Map-based Multi-Policy Reinforcement Learning: Enhancing Adaptability of Robots by Deep Reinforcement Learning
• Repairing Neural Networks for Safety in Robotic Systems using Predictive Models
• Survivable Robotic Control through Guided Bayesian Policy Search with Deep Reinforcement Learning
• Towards semi-episodic learning for robot damage recovery