Evaluation Results of Benchmarking Actor-Critic Deep Reinforcement Learning Algorithms for Robotics Control with Action Constraints

Citation Author(s):
Kazumi
Kasaura
OMRON SINIC X Corporation
Shuwa
Miura
University of Massachusetts Amherst
Tadashi
Kozuno
OMRON SINIC X Corporation
Ryo
Yonetani
OMRON SINIC X Corporation
Kenta
Hoshino
Kyoto University
Yohei
Hosoe
Kyoto University
Submitted by:
Kazumi Kasaura
Last updated:
Mon, 06/12/2023 - 21:54
DOI:
10.21227/qwd2-jd89
Data Format:
Research Article Link:
License:
3
2 ratings - Please login to submit your rating.

Abstract 

This study presents a benchmark for evaluating action-constrained reinforcement learning (RL) algorithms. In action-constrained RL, each action taken by the learning system must comply with certain constraints. These constraints are crucial for ensuring the feasibility and safety of actions in real-world systems. We evaluate existing algorithms and their novel variants across multiple robotics control environments, encompassing multiple action constraint types. Our evaluation provides the first in-depth perspective of the field, revealing surprising insights, including the effectiveness of a straight- forward baseline approach. The benchmark problems and as- sociated code utilized in our experiments are made avail- able online at github.com/omron-sinicx/action-constrained-RL- benchmark for further research and development.

Instructions: 

For task A and algorithm B, the result of evaluation during training with seed S is stored in 'result_data/logs-A/B-S/evaluations.npz'.

For task A and algorithm B, the result of the evaluation script (evaluation.py in the github repository) is stored in 'result_data/logs-A/B/result.npy'. It contains the result of evaluations for all 10 seeds.

All .npz files and .npy files can be opened by python scripts with numpy.

For any quastion, please contact: kazumi.kasaura@sinicx.com.

Dataset Files

    Files have not been uploaded for this dataset