I am a research scientist, building humanoid robots, at Tangible.
Previously, I was an undergraduate researcher mentored by Anthony Liang, Yigit Korkmaz and Prof. Erdem Bıyık (LiRA Lab). I work to create safe, responsible AI systems for the real world.
Publications
RHODES: Reducing Human Oversight via Disagreement and Exploration for Safe Reinforcement Learning
Jaiv Doshi*, Anthony Liang*, Yigit Korkmaz, Erdem Bıyık
In Submission to ICML 2025
Projects
Generative Social Agents with Evolving Personas
At the Ahmanson Lab, I built an AI system that simulates human-like social media interactions using an adaptive context-based decision making framework. The system enabled AI agents to maintain evolving personas, build relationships, and respond contextually to conversations in real-time.
Adaptive Robotic Exploration for Haptic Property Recognition
In our final project, we developed a multimodal perception system that helps robots recognize object properties by combining haptic and auditory feedback. We built a ResNet-based neural network along with an adaptive exploration strategy that selects the most informative physical interactions based on expected information gain.