Oliver Wang

M.S. student, Electrical Engineering, UCLA · NSF Graduate Research Fellow

I’m a master’s student at UCLA, working with Prof. Mani Srivastava in the Networked and Embedded Systems Lab (NESL) and with Prof. Abeer Alwan in the Speech Processing and Auditory Perception Lab (SPAPL). I finished my B.S. in Electrical Engineering at UCLA in June 2026.

I’m interested in robots and agents that keep getting better over long-term deployment. My current work examines how home robots should update their memories as household routines change, and whether model confidence helps identify incorrect predictions. I’m now testing these ideas on a Unitree Go2 quadruped, where the real world brings failures that never showed up in simulation. I also lead the research direction of a project with Sandia National Laboratories on detecting urban events like wildfires and traffic accidents from cameras, sensors, and social media. Meanwhile, I’ve also enjoyed working with time-series, such as identifying whether a given time-series foundation model is suitable for a dataset.

Interests

• Memory that revises its beliefs as environments and human routines change
• Uncertainty that stays meaningful under distribution shift
• Agents that diagnose and learn from their own failures
• Knowing when to trust a model (reliability signals for foundation models)

Selected publications

• CoRL-W
  How Should a Home Robot Update Its Memories When Human Routines Change?

  O. Wang, M. Srivastava

  CoRL 2026 Workshop on Memory for Robot Foundation Models (RoboMem) accepted; companion versions at the CoRL 2026 workshops on Modeling Uncertainty in Robotic World Models and on Continually Self-Improving Robots

  LLM-written memory summaries often forget what they learned when an old routine comes back. Keeping hypotheses about routines, backed by raw observations, makes predictions hold up better under these shifts.

• AAAI-W
  Spectral Predictability as a Fast Reliability Indicator for Time Series Forecasting Model Selection

  O. Wang, P. Quan, K. Yang, M. Srivastava

  AAAI 2026 Workshop on AI for Time Series (AI4TS) oral

  A score computed in seconds tells you whether a forecasting foundation model will beat a small, cheap model on your data.

  arXivcode
• AROB
  Robots Reading Recipes: Large Language Models as Translators Between Humans and Machines

  O. Wang, G. Cheng, L. Caspar, A. Yokota, M. Khosravy, O. Witkowski

  Artificial Life and Robotics, 30(3), 2025

  Turning written cocktail recipes into robot commands with an off-the-shelf LLM and no fine-tuning.

  paper

All publications →

News