Potential-based Reward Shaping Research Articles

Multi-agent credit assignment in stochastic resource management games

AbstractMulti-agent systems (MASs) are a form of distributed intelligence, where multiple autonomous agents act in a common environment. Numerous complex, real world systems have been successfully optimized using multi-agent reinforcement learning (MARL) in conjunction with the MAS framework. In MARL agents learn by maximizing a scalar reward signal from the environment, and thus the design of the reward function directly affects the policies learned. In this work, we address the issue of appropriate multi-agent credit assignment in stochastic resource management games. We propose two new stochastic games to serve as testbeds for MARL research into resource management problems: the tragic commons domain and the shepherd problem domain. Our empirical work evaluates the performance of two commonly used reward shaping techniques: potential-based reward shaping and difference rewards. Experimental results demonstrate that systems using appropriate reward shaping techniques for multi-agent credit assignment can achieve near-optimal performance in stochastic resource management games, outperforming systems learning using unshaped local or global evaluations. We also present the first empirical investigations into the effect of expressing the same heuristic knowledge in state- or action-based formats, therefore developing insights into the design of multi-agent potential functions that will inform future work.

Plan-based reward shaping for multi-agent reinforcement learning

AbstractRecent theoretical results have justified the use of potential-based reward shaping as a way to improve the performance of multi-agent reinforcement learning (MARL). However, the question remains of how to generate a useful potential function.Previous research demonstrated the use of STRIPS operator knowledge to automatically generate a potential function for single-agent reinforcement learning. Following up on this work, we investigate the use of STRIPS planning knowledge in the context of MARL.Our results show that a potential function based on joint or individual plan knowledge can significantly improve MARL performance compared with no shaping. In addition, we investigate the limitations of individual plan knowledge as a source of reward shaping in cases where the combination of individual agent plans causes conflict.

Reinforcement-learning based dialogue system for human–robot interactions with socially-inspired rewards

This paper investigates some conditions under which polarized user appraisals gathered throughout the course of a vocal interaction between a machine and a human can be integrated in a reinforcement learning-based dialogue manager. More specifically, we discuss how this information can be cast into socially-inspired rewards for speeding up the policy optimisation for both efficient task completion and user adaptation in an online learning setting. For this purpose a potential-based reward shaping method is combined with a sample efficient reinforcement learning algorithm to offer a principled framework to cope with these potentially noisy interim rewards. The proposed scheme will greatly facilitate the system's development by allowing the designer to teach his system through explicit positive/negative feedbacks given as hints about task progress, in the early stage of training. At a later stage, the approach will be used as a way to ease the adaptation of the dialogue policy to specific user profiles. Experiments carried out using a state-of-the-art goal-oriented dialogue management framework, the Hidden Information State (HIS), support our claims in two configurations: firstly, with a user simulator in the tourist information domain (and thus simulated appraisals), and secondly, in the context of man–robot dialogue with real user trials.

Potential-based reward shaping for finite horizon online POMDP planning

In this paper, we address the problem of suboptimal behavior during online partially observable Markov decision process (POMDP) planning caused by time constraints on planning. Taking inspiration from the related field of reinforcement learning (RL), our solution is to shape the agent's reward function in order to lead the agent to large future rewards without having to spend as much time explicitly estimating cumulative future rewards, enabling the agent to save time to improve the breadth planning and build higher quality plans. Specifically, we extend potential-based reward shaping (PBRS) from RL to online POMDP planning. In our extension, information about belief states is added to the function optimized by the agent during planning. This information provides hints of where the agent might find high future rewards beyond its planning horizon, and thus achieve greater cumulative rewards. We develop novel potential functions measuring information useful to agent metareasoning in POMDPs (reflecting on agent knowledge and/or histories of experience with the environment), theoretically prove several important properties and benefits of using PBRS for online POMDP planning, and empirically demonstrate these results in a range of classic benchmark POMDP planning problems.

Expressing Arbitrary Reward Functions as Potential-Based Advice

Effectively incorporating external advice is an important problem in reinforcement learning, especially as it moves into the real world. Potential-based reward shaping is a way to provide the agent with a specific form of additional reward, with the guarantee of policy invariance. In this work we give a novel way to incorporate an arbitrary reward function with the same guarantee, by implicitly translating it into the specific form of dynamic advice potentials, which are maintained as an auxiliary value function learnt at the same time. We show that advice provided in this way captures the input reward function in expectation, and demonstrate its efficacy empirically.

AN EMPIRICAL STUDY OF POTENTIAL-BASED REWARD SHAPING AND ADVICE IN COMPLEX, MULTI-AGENT SYSTEMS

This paper investigates the impact of reward shaping in multi-agent reinforcement learning as a way to incorporate domain knowledge about good strategies. In theory, potential-based reward shaping does not alter the Nash Equilibria of a stochastic game, only the exploration of the shaped agent. We demonstrate empirically the performance of reward shaping in two problem domains within the context of RoboCup KeepAway by designing three reward shaping schemes, encouraging specific behaviour such as keeping a minimum distance from other players on the same team and taking on specific roles. The results illustrate that reward shaping with multiple, simultaneous learning agents can reduce the time needed to learn a suitable policy and can alter the final group performance.

Online learning of shaping rewards in reinforcement learning

Potential-based reward shaping has been shown to be a powerful method to improve the convergence rate of reinforcement learning agents. It is a flexible technique to incorporate background knowledge into temporal-difference learning in a principled way. However, the question remains of how to compute the potential function which is used to shape the reward that is given to the learning agent. In this paper, we show how, in the absence of knowledge to define the potential function manually, this function can be learned online in parallel with the actual reinforcement learning process. Two cases are considered. The first solution which is based on the multi-grid discretisation is designed for model-free reinforcement learning. In the second case, the approach for the prototypical model-based R-max algorithm is proposed. It learns the potential function using the free space assumption about the transitions in the environment. Two novel algorithms are presented and evaluated.