Dual REPS: A Generalization of Relative Entropy Policy Search Exploiting Bad Experiences

• Published in: IEEE transactions on robotics 2017-08, Vol.33 (4), p.978-985
• Main Authors: Colome, Adria, Torras, Carme
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial intelligence; Classificació INSPEC; Clustering algorithms; Clusters; Computer simulation; Cybernetics; Entropy; Experiments; Gaussian distribution; Informàtica; Learning (artificial intelligence); Low-reward data reuse; Machine learning; Maxima; Optimization; policy search (PS); Probability distribution; Randomness; reinforcement learning (RL); relative entropy policy search (REPS); Robots; Robòtica; Searching; Statistical analysis; Statistical methods; Àrees temàtiques de la UPC
• ISSN: 1552-3098; 1941-0468
• DOI: 10.1109/TRO.2017.2679202

Policy search (PS) algorithms are widely used for their simplicity and effectiveness in finding solutions for robotic problems. However, most current PS algorithms derive policies by statistically fitting the data from the best experiments only. This means that experiments yielding a poor performance are usually discarded or given too little influence on the policy update. In this paper, we propose a generalization of the relative entropy policy search (REPS) algorithm that takes bad experiences into consideration when computing a policy. The proposed approach, named dual REPS (DREPS) following the philosophical interpretation of the duality between good and bad, finds clusters of experimental data yielding a poor behavior and adds them to the optimization problem as a repulsive constraint. Thus, considering that there is a duality between good and bad data samples, both are taken into account in the stochastic search for a policy. Additionally, a cluster with the best samples may be included as an attractor to enforce faster convergence to a single optimal solution in multimodal problems. We first tested our proposed approach in a simulated reinforcement learning setting and found that DREPS considerably speeds up the learning process, especially during the early optimization steps and in cases where other approaches get trapped in between several alternative maxima. Further experiments in which a real robot had to learn a task with a multimodal reward function confirm the advantages of our proposed approach with respect to REPS.