Hi there š Iām Jiexin (Judy) Wang ;)
I am a robotics and AI researcher interested in one fundamental question:
How does intelligence emerge?
My research aims to understand how intelligent behavior develops in humans and animals through interaction with the real world, and how similar computational principles can be realized in autonomous robotic systems. I see robotics not only as an application of AI, but also as a scientific tool for studying intelligence.
I am currently seeking research opportunities in Robot Learning, Embodied AI, and Autonomous Intelligent Systems.
Previously, I was a postdoctoral researcher working with Dr. Eiji Uchibe in the Department of Brain Robot Interface at ATR Computational Neuroscience Laboratories.
I received my PhD under the supervisions of Prof. Shin Ishii at the Integrated Systems Biology Lab, Grad School of Informatics, Kyoto University, and Prof. Kenji Doya in the Adaptive Systems Group, Neural Computation Unit, Okinawa Institute of Science and Technology (OIST).
Outside research, I enjoy climbing and spending time in nature.
Embodied Behavior
My PhD research explored the first step toward intelligent behavior:
How can autonomous behaviors be acquired?
I built an affordable and sustainable smartphone-based robotic learning platform from scratch, covering hardware, sensing, control and reinforcement learning.
Using this platform, I developed efficient policy-search algorithms for investigating a range of embodied behaviors, including self-standing, vision-based navigation, foraging and genetic information exchange between multiple agents (mating). Together, these behaviors enabled autonomous agents to interact and organize into simple multi-agent societies inspired by living creatures.
These experiences convinced me that intelligent behavior is best studied in embodied systems, where interaction with the physical world plays an essential role in shaping learning and behavior.
Motivation and Reinforcement Learning
My postdoctoral research explored another aspect of intelligent behavior:
How do internal motivational systems shape behavior?
Inspired by neuroscience, I developed reinforcement learning frameworks that explicitly separate reward and punishment into interacting learning processes. I view these as multiple internal motivational systems that cooperate within a single agent.
A complete deep reinforcement learning pipeline was built upon autonomous mobile robots, integrating multi-modal perception, real-time robot control and policy learning in real-world environments.
This work resulted in a series of deep reinforcement learning algorithms, including DMP, softDMP and klDMP, to tackle several long-standing issues in reinforcement learning, including reward sparsity, exploration, safety, robustness and sample efficiency.
Towards Autonomous Intelligence
More recently, my research has expanded beyond classical reinforcement learning. I am exploring what additional computational mechanisms may be required for autonomous intelligence in embodied agents. Related topics include intrinsic motivation, curriculum learning, life-long learning, world models, and foundation models for robotics.