机器人ICLR 20252025

SimplerEnv: Evaluating Real-World Robot Policies in Simulation

Xuanlin Li, Kyle Hsu, Jiayuan Gu, Karl Pertsch, Chelsea FinnStanford University, UC Berkeley

摘要

We present SimplerEnv, a simulation-based evaluation framework for assessing real-world robot policy performance. SimplerEnv provides photorealistic environments with accurate physics to benchmark manipulation policies, revealing critical gaps between simulated and real-world performance.

robot evaluationsimulation benchmarkpolicy assessmentphotorealisticmanipulation

技术细节

数据集
SimplerEnv BenchmarkOpen X-EmbodimentRT-1 Data
测试机器人
FR3WidowX-250
模型骨架

Evaluation Framework (multiple policy baselines)

编码器

ViT + Proprioceptive Encoders

解码器

Policy Decoders (RT-1, Octo, OpenVLA variants)

京ICP备2026064258号-1