Leveraging Mutual Information for Asymmetric Learning under Partial Observability
Hai Nguyen, Long Dinh, Robert Platt, Christopher Amato.
Conference on Robot Learning (CoRL), 2024
We used mutual information between state (assumed available only during training) and history to compute intrinsic rewards
to encourage actions that create a history that can explain the state, hence, improving information gathering.
|
|
Symmetry-aware Reinforcement Learning for Robotic Assembly under Partial Observability with a Soft Wrist
Hai Nguyen, Tadashi Kozuno, Cristian C. Beltran-Hernandez, Masashi Hamaya.
IEEE International Conference on Robotics and Automation (ICRA), 2024
We used domain symmetry to learn the Peg-In-Hole task (using F/T feedback and 50 episodes of human demonstration) directly on a soft UR5e robot in two hours.
[code]  / 
[site + videos]
|
|
On-Robot Bayesian Reinforcement Learning for POMDPs
Hai Nguyen, Sammie Katt, Yuchen Xiao, Christopher Amato.
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2023
We used the framework of Bayesian Adaptive RL and mixed-observability assumption to quickly learn a planning-based policy directly on hardware.
[robot video]  / 
[poster]
|
|
Equivariant Reinforcement Learning under Partial Observability
Hai Nguyen, Andrea Baisero, David Klee, Dian Wang, Robert Platt, Christopher Amato.
Conference on Robot Learning (CoRL), 2023
We leveraged the symmetry in SE(2) using spherical CNNs for memory-based agents to solve a class of symmetric POMDPs.
[code]  / 
[site + videos]
|
|
Leveraging Fully Observable Policies for Learning under Partial Observability
Hai Nguyen, Andrea Baisero, Dian Wang, Christopher Amato, Robert Platt.
Conference on Robot Learning (CoRL), 2022
We proposed a method to learn POMDP policies under the guidance of MDP expert policies.
[code]  / 
[site + videos]
|
|
Hierarchical Reinforcement Learning under Mixed Observability
Hai Nguyen*, Zhihan Yang*, Andrea Baisero, Xiao Ma, Robert Platt, Christopher Amato.
International Workshop on the Algorithmic Foundations of Robotics (WAFR), 2022
We proposed a hierarchical agent addressing a class of POMDPs that only needs to works with partial observability on the top level.
[site + videos]
|
|
Belief-Grounded Networks for Accelerated Robot Learning under Partial Observability
Hai Nguyen*, Brett Daley*, Xinchao Song, Christopher Amato, Robert Platt.
Conference on Robot Learning (CoRL), 2020
We proposed auxiliary losses to leverage belief states during training to improve the learning efficiency of actor-critic agents.
[code]  / 
[site + videos]
|
|
|
CoRL (2024),
ICRA (2021, 2024),
RAL (2022, 2023),
IROS (2022, 2023),
RO-MAN (2022)
|