Finally, it’s my pleasure to announce that our paper with the title “Multi-objective Model-based Policy Search for Data-efficient Learning with Sparse Rewards” has been accepted in Conference on Robot Learning (CoRL) - 2018 which will be held on October 29th-31st, 2018, in Zürich, Switzerland. This time the acceptance rate in CoRL was 31%. Only 75 out of 237 papers have been accepted.
In this paper we proposed a novel model-based policy search algorithm called Multi-DEX to deal with sparse reward scenarios in a data-efficient manner. Multi-DEX is capable of learning new policies under sparse reward landscape where many state of the art algorithms like PILCO and Black-Drops perform poorly. Compared to model free policy gradient based approaches such as TRPO, TRPO with VIME, GEP-PG out proposed algorithm is abe able to solve tasks with sparse rewards a number of magnitude fewer interaction time with the system while learning the system dynamics form scratch.
The arxiv version of the paper can be found here: Multi-objective Model-based Policy Search for Data-efficient Learning with Sparse Rewards