Safety Correction from Baseline: Towards the Risk-aware Policy in Robotics via Dual-agent Reinforcement Learning.
Linrui ZhangZichen YanLi ShenShoujie LiXueqian WangDacheng TaoPublished in: CoRR (2022)
Keyphrases
- reinforcement learning
- action selection
- reward function
- markov decision process
- agent learns
- state action
- agent receives
- partially observable
- optimal policy
- multi agent
- action space
- policy search
- decision making
- markov decision processes
- state space
- learning agent
- reinforcement learning algorithms
- intelligent agents
- function approximators
- function approximation
- selective perception
- single agent
- computer vision
- temporal difference
- control policy
- actor critic
- markov decision problems
- multi agent environments
- artificial intelligence
- expected reward
- inverse reinforcement learning
- multi agent systems
- partially observable environments
- policy iteration
- partially observable markov decision process
- infinite horizon
- control policies
- dynamic programming
- multiagent systems
- software agents
- robot control
- autonomous learning
- human users
- reinforcement learning problems
- reward signal
- state and action spaces
- policy gradient
- learning capabilities
- reinforcement learning methods
- learning algorithm
- autonomous agents
- risk management
- risk assessment
- agent model
- multiple agents
- belief state
- continuous state
- model free
- state abstraction
- exploration strategy
- decision theoretic
- expected utility
- agent architecture
- optimal control
- reward shaping
- machine learning
- partially observable domains
- policy evaluation
- dynamic environments