Robot Fine-Tuning Made Easy: Pre-Training Rewards and Policies for Autonomous Real-World Reinforcement Learning.
Jingyun YangMax Sobol MarkBrandon VuArchit SharmaJeannette BohgChelsea FinnPublished in: CoRR (2023)
Keyphrases
- fine tuning
- reinforcement learning
- autonomous learning
- real world
- real robot
- optimal policy
- fine tuned
- reward function
- robotic systems
- markov decision processes
- function approximation
- control policies
- policy search
- perceptual aliasing
- fine tune
- robot behavior
- control policy
- robot control
- semi autonomous
- total reward
- mobile robot
- partially observable markov decision processes
- viable alternative
- reinforcement learning algorithms
- supervised learning
- state space
- markov decision process
- markov decision problems
- service robots
- real environment
- autonomous vehicles
- autonomous navigation
- decentralized control
- hierarchical reinforcement learning
- multi robot
- fitted q iteration
- human robot interaction
- data sets
- vision system
- expected reward
- robot navigation
- search and rescue
- infinite horizon
- reward shaping
- long run
- motor learning
- learning algorithm
- multi agent
- training set
- dynamic programming
- model free
- robot arm
- motor skills
- decision problems
- average cost
- dynamic environments
- internal representations
- average reward
- macro actions
- machine learning
- finite state
- autonomous robots