Thinh Doan
University of Texas at Austin, TX·Engineering·Early-career
When many robots must work together in real time, such as drones scanning a wildfire or searching for a lost hiker, each one has to learn good decisions from experience. Reinforcement learning is a data-driven way for machines to learn tasks by trial and error, interacting directly with their surroundings.
- Fields
Engineering
Computer Science
Mathematics & Statistics
- Topics
reinforcement learning
multi-agent systems
distributed learning
- Grants
- 2 known grants, through Feb 2029
What they work on
- Develop the theory of two-time-scale stochastic approximation for non-monotone systems, using extrapolation from optimization and singular perturbation from control to keep these noisy iterative methods stable, and prove finite-time complexity bounds.
- Design distributed multi-time-scale reinforcement learning methods that let robot teams cooperate in real time despite changing numbers of agents, failures, limited communication and adversaries.
- Evaluate the resilient distributed learning methods through simulations and field experiments of multi-robot navigation.
- Determine under what conditions learning multiple reinforcement learning tasks jointly needs less data and computation than learning each alone, using tools from online learning, compressed sensing and stochastic modeling.
- Learn the similarity structure between tasks from offline datasets and use it to adapt quickly to new tasks online, while addressing catastrophic forgetting of earlier tasks.
More about the research
It works well in principle but needs huge amounts of data and computing power. It can also become unstable when errors, failures or adversaries appear. The research here builds mathematical foundations to make this kind of learning faster, cheaper in data and more robust for teams of agents. Both projects are tested in simulations of multi-robot navigation.
Why it matters
Teams of autonomous robots and drones could help with wildfire monitoring, search-and-rescue and manufacturing, but current learning methods need too much data and computation and can fail when robots break down or face interference. More data-efficient and fault-tolerant learning methods could make such multi-robot systems practical in settings where data, computing power and communication are limited.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Ioannis Kantaros
Washington University in St Louis, MO·Engineering·Early-career
Train multi-agent reinforcement learning (MARL) control policies for robot teams and mathematically certify their robustness against adversarial, high-dimensional sensor inputs, testing them on aerial and ground robot testbeds and photorealistic simulators.
Yuxin Chen
University of Pennsylvania, PA·Mathematics & Statistics
Characterize the fundamental limits on sample and computational complexity of reinforcement learning, meaning how much data and computing an agent needs to learn good decisions in an unknown environment.
Vijay Subramanian
University of Michigan, MI·Engineering
Develop robust reinforcement learning algorithms that use model-class knowledge, future predictions and off-policy methods to cope with mismatches and limited data from simulators or offline records.
The school
University of Texas at Austin: 49 labs on PI Pages, $1.1B a year on research, 871 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.