NeoRL: A Near Real-World Benchmark for Offline Reinforcement Learning

2021/02/01 by Rongjun Qin, Qin, Rongjun, Songyi Gao +13 · 4 citations
Computer Science · Engineering · #Artificial Intelligence (cs.AI) #Data Stream Mining Techniques #FOS: Computer and information sciences #Machine Learning (cs.LG) #Reinforcement Learning in Robotics #Smart Grid Energy Management #cs.AI #cs.LG

paper · pdf · doi:10.48550/arxiv.2102.00714

openalex publication_date 2021/02/01 · arxiv created 2021/02/08 · arxiv updated 2021/02/09 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/29

Abstract

Offline reinforcement learning (RL) aims at learning a good policy from a batch of collected data, without extra interactions with the environment during training. However, current offline RL benchmarks commonly have a large reality gap, because they involve large datasets collected by highly exploratory policies, and the trained policy is directly evaluated in the environment. In real-world situations, running a highly exploratory policy is prohibited to ensure system safety, the data is commonly very limited, and a trained policy should be well validated before deployment. In this paper, we present a near real-world offline RL benchmark, named NeoRL, which contains datasets from various domains with controlled sizes, and extra test datasets for policy validation. We evaluate existing offline RL algorithms on NeoRL and argue that the performance of a policy should also be compared with the deterministic version of the behavior policy, instead of the dataset reward. The empirical results demonstrate that the tested offline RL algorithms become less competitive to the deterministic policy on many datasets, and the offline policy evaluation hardly helps. The NeoRL suit can be found at http://polixir.ai/research/neorl. We hope this work will shed some light on future research and draw more attention when deploying RL in real-world systems.

Citations

• Playing Atari with Deep Reinforcement Learning
• DeepMind Control Suite
• Benchmarking Batch Deep Reinforcement Learning Algorithms
• Empirical Study of Off-Policy Policy Evaluation for Reinforcement Learning
• Behavior Regularized Offline Reinforcement Learning
• An empirical investigation of the challenges of real-world reinforcement learning
• D4RL: Datasets for Deep Data-Driven Reinforcement Learning
• Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems
• Relay Policy Learning: Solving Long-Horizon Tasks via Imitation and Reinforcement Learning
• PLAS: Latent Action Space for Offline Reinforcement Learning

Cited by

• Long-Horizon Model-Based Offline Reinforcement Learning Without Explicit Conservatism
• PyTupli: A Scalable Infrastructure for Collaborative Offline Reinforcement Learning Projects
• ImagineBench: Evaluating Reinforcement Learning with Large Language Model Rollouts
• Towards Automatic Actor-Critic Solutions to Continuous Control
• VIPO: Value Function Inconsistency Penalized Offline Reinforcement Learning

Related

• NeoRL-2: Near Real-World Benchmarks for Offline Reinforcement Learning with Extended Realistic Scenarios
• Improving and Benchmarking Offline Reinforcement Learning Algorithms
• Offline RL Without Off-Policy Evaluation
• Don't Change the Algorithm, Change the Data: Exploratory Data for Offline Reinforcement Learning
• D4RL: Datasets for Deep Data-Driven Reinforcement Learning