Learning Motor Skills: From Algorithms to Robot Experiments by Jens Kober, Jan Peters

By Jens Kober, Jan Peters

This publication provides the cutting-edge in reinforcement studying utilized to robotics either by way of novel algorithms and purposes. It discusses fresh methods that let robots to benefit motor.

skills and offers initiatives that have to bear in mind the dynamic habit of the robotic and its surroundings, the place a kinematic stream plan isn't really enough. The e-book illustrates a style that learns to generalize parameterized motor plans that is bought by means of imitation or reinforcement studying, via adapting a small set of worldwide parameters and acceptable kernel-based reinforcement studying algorithms. The offered functions discover hugely dynamic initiatives and convey a truly effective studying procedure. All proposed techniques were widely demonstrated with benchmarks projects, in simulation and on actual robots. those initiatives correspond to activities and video games however the offered thoughts also are appropriate to extra mundane responsibilities. The e-book is predicated at the first author’s doctoral thesis, which received the 2013 EURON Georges Giralt PhD Award.

Show description

Read Online or Download Learning Motor Skills: From Algorithms to Robot Experiments PDF

Similar robotics & automation books

Parallel Robots

Parallel robots are closed-loop mechanisms providing excellent performances by way of accuracy, stress and skill to govern huge quite a bit. Parallel robots were utilized in a good number of functions starting from astronomy to flight simulators and have gotten more and more well known within the box of machine-tool undefined.

Advanced Neural Network-Based Computational Schemes for Robust Fault Diagnosis

The current booklet is dedicated to difficulties of variation of man-made neural networks to powerful fault analysis schemes. It provides neural networks-based modelling and estimation thoughts used for designing strong fault prognosis schemes for non-linear dynamic platforms. part of the e-book specializes in primary matters similar to architectures of dynamic neural networks, equipment for designing of neural networks and fault analysis schemes in addition to the significance of robustness.

Optimal and Robust Estimation: With an Introduction to Stochastic Control Theory, Second Edition

Greater than a decade in the past, world-renowned regulate platforms authority Frank L. Lewis brought what could develop into a typical textbook on estimation, below the identify optimum Estimation, utilized in best universities in the course of the global. The time has come for a brand new variation of this vintage textual content, and Lewis enlisted the help of finished specialists to carry the e-book thoroughly modern with the estimation tools riding trendy high-performance structures.

Extra resources for Learning Motor Skills: From Algorithms to Robot Experiments

Sample text

Using the state-action value function Q∗ (s, a) instead of the value function V ∗ (s) π ∗ (s) = arg max (Q∗ (s, a)) , a avoids having to calculate the weighted sum over the successor states, and hence no knowledge of the transition function is required. A wide variety of methods of value function based reinforcement learning algorithms that attempt to estimate V ∗ (s) or Q∗ (s, a) have been developed and can be split mainly into three classes: (i) dynamic programmingbased optimal control approaches such as policy iteration or value iteration, (ii) rollout-based Monte Carlo methods and (iii) temporal difference methods such as TD(λ) (Temporal Difference learning), Q-learning, and SARSA (State-Action-Reward-State-Action).

An action taken does not have to have an immediate effect on the reward but can also influence a reward in the distant future. The difficulty in assigning credit for rewards is directly related to the horizon or mixing time of the problem. It also increases with the dimensionality of the actions as not all parts of the action may contribute equally. The classical reinforcement learning setup is a MDP where additionally to the states S, actions A, and rewards R we also have transition probabilities T (s , a, s).

In general in robotics, we may only be able to find some approximate notion of state. Different types of reward functions are commonly used, including rewards depending only on the current state R = R(s), rewards depending on the current state and action R = R(s, a), and rewards including the transitions R = R(s , a, s). Most of the theoretical guarantees only hold if the problem adheres to a Markov structure, however in practice, many approaches work very well for many problems that do not fulfill this requirement.

Download PDF sample

Rated 4.86 of 5 – based on 32 votes