Hey! I’m Richard, a second-year undergrad at Caltech studying Computer Science and minoring in Mathematics and Control & Dynamical Systems (CDS).
My interest is in building scalable models for solving hard problems. I’m particularly interested in applications to self-driving vehicles and intelligent robotics, specifically through spatial reasoning, large language models, model predictive control, and perception/vision.
Currently, I’m improving VLM reasoning and spatial awareness in Glab. I’m also working on LLM pre and post-training to prove Olympiad-level (and beyond) inequalities under Prof. Tony Yue Yu at Caltech. I’ve previously worked under Dr. Alec Reed at CU Boulder’s Autonomous Robotics Lab on predictive vehicle dynamics.
Last summer, I worked on software development and operations research at Commerzbank in New York City. I’m in Seattle this summer interning at Amazon!
If you’d like to chat, please reach out at rhoffman@caltech.edu.
Recent News
last updated: June 2025
-
Adversarial attacks on stochastic bandits often rely on unrealistic assumptions like unrestricted, per-round reward manipulation. We introduce a more practical threat model, Fake Data Injection, where the attacker injects a limited number of bounded fake feedback samples. We demonstrate both theoretically and empirically that Practical Adversarial Attacks on Stochastic Bandits via Fake Data Injection can effectively mislead popular algorithms like UCB and Thompson Sampling with minimal attack cost.
-
We explore a novel neural population code method to accurately estimate object orientation. I’m excited to see how Object-Pose Estimation With Neural Population Codes can be scaled to enhance robot perception and improve autonomous vehicle driving.