About me
Welcome to Haruki Nishimura’s website! I am a Senior Research Scientist at Toyota Research Institute, focused on Trustworthy Machine Learning for Embodied Systems under Uncertainty. Prior to joining TRI, I was a Ph.D. student at Stanford University, advised by Prof. Mac Schwager in Multi-Robot Systems Lab. I am passionate about making embodied autonomous systems safer and more robust so people can truly trust and rely on them in the AI-assisted society of the near future. Since 2018, I have developed advanced trajectory planning and learning algorithms for autonomous robots that must interactively operate without failure under perceptual, dynamic, and model uncertainty. Those algorithms leverage tools from numerical optimization, optimal decision/control theory, probabilistic inference, reinforcement learning, and statistics.
I received my Ph.D. degree and M.S. degree from Stanford University in 2021 and 2017, respectively, and B.Eng. from the University of Tokyo in 2015, all in Aeronautics and Astronautics. At UTokyo, I was researching feature-based aerial SLAM algorithms for life-logging applications, under the supervision of Prof. Takehisa Yairi at the Research Center for Advanced Science and Technology.
Apart from research activities, I am passionate about stargazing, traveling, as well as science/engineering education.

Recent News
September 27, 2025: Workshop talk at CoRL
I gave an invited talk at the 2nd Workshop on Safe and Robust Robot Learning for Operation in the Real World (SAFE-ROL). I sincerely thank the organizers for the opportunity to speak at the workshop.
September 18, 2025: Two Papers accepted at NeurIPS!
Two university collaboration papers entitled STITCH-OPE: Trajectory Stitching with Guided Diffusion for Off-Policy Evaluation and SAFE: Scalable Failure Estimation for Vision-Language-Action Models have been accepted at Neural Information Processing Systems (NeurIPS) 2025. STITCH-OPE is going to be presented as a spotlight (acceptance rate: 3.19%) and SAFE is a poster (acceptance rate: 24.52%). We look forward to presenting our work in December!
August 28, 2025: Seminar at UTokyo
I gave an invited seminar for the GVLab Seminar Series at the University of Tokyo in Japan. I sincerely thank Prof. Gentiane Venture for the kind invitation.
August 1, 2025: Paper accepted at CoRL!
A university collaboration paper entitled CUPID: Curating Data your Robot Loves with Influence Functions has been accepted at Conferene on Robot Learning (CoRL) 2025. The acceptance rate was 35.77%. We look forward to presenting in our work in Seoul, South Korea in September!
June 21, 2025: 1st Workshop on Robot Evaluation at RSS
With co-organizers from both academia and industry, we held the 1st workshop on Robot Evaluation for the Real World at RSS 2025. Thank you to those who attended and/or contributed to the workshop on such an important and timely topic!
June 21, 2025: Workshop talk at RSS
I gave an invited talk at the RSS 2025 Workshop on Reliable Robotics: Safety and Security in the Face of Generative AI. I sincerely thank the organizers for having me at the workshop.
April 16, 2025: Guest lecture at CMU
I gave a guest lecture in the Embodied Artificial Intelligence Safety class at CMU Robotics Institute on the effective use of statistical methods for trustworthy policy evaluation. I sincerely thank Prof. Andrea Bajcsy for the kind invitation.
April 10, 2025: Two papers accepted at RSS!
Two papers entitled Is Your Imitation Learning Policy Better than Mine? Policy Comparison with Near-Optional Stopping and Can We Detect Failures without Failure Data? Uncertainty-Aware Runtime Failure Detection for Imitation Learning Policies have been accepted at Robotics: Science and Systems (RSS) 2025. The acceptance rate was 27%. We look forward to presenting our work in Los Angeles, USA in June!
August 19, 2024: Paper accepted into RA-L!
Our latest work entitled How Generalizable is My Behavior Cloning Policy? A Statistical Approach to Trustworthy Performance Evaluation has been accepted into IEEE Robotis and Automation Letters (RA-L)! Check out the project website for details.
April 17, 2024: Paper featured in Nikkei Robotics
Our NeurIPS paper on Policy Customization has been featured in the May 2024 release of Nikkei Robotics, a monthly Japanese magazine on Robotics and AI.
December 11, 2023: Recognized as a Top Reviewer at NeurIPS 2023
It is such an honor to be recognized as a Top Reviewer at NeurIPS 2023!
June 15, 2023: Received Best Paper Award!
Our L4DC paper won the Best Paper Award!
April 11, 2023: Paper accepted (oral) at L4DC!
Our latest work entitled In-Distribution Barrier Functions: Self-Supervised Policy Filters that Avoid Out-of-Distribution States has been accepted for an oral presentation at the 5th Annual Learning for Dynamics & Control Conference (L4DC). The acceptance rate for oral presentations was 9.6%. We look forward to presenting our work in Philadelphia, USA in June!
September 10, 2022: Paper accepted (oral) at CoRL!
Our latest work on risk-aware trajectory forecasting has been accepted for an oral presentation at the 2022 Conference on Robot Learning. The acceptance rate for oral presentations was 6.5%. The arXiv version is available here. We look forward to presenting our work in Auckland, New Zealand in December!