Joohwan Seo

News

• I will present a paper, titled "A Comparison between Lie Group- and Lie Algebra-based Potential Functions for Geometric Impedance Control", at ACC 2024, Toronto, Canada, July 10-12.
  Also serving as a Session chair in Autonomous Robots!

• I will present a paper, titled "Contact-rich SE(3) Equivariant Robot Manipulation Task Learning via Geometric Impedance Control", at IROS 2024, Abudhabi, UAE, Oct 14-18.

Current Research Interests / Open Questions

• Geometric Impedance Control (GIC)
  - Geometric Impedance Control as an equivariant dynamic control law on SE(3)
  - Utilization of Geometrically Consistent Error Vectors (GCEV)
  - Learning impedance gains for GIC in application to robotic assembly tasks

• Dealing with Vision & Pointcloud input
  - There are many geometric approaches
  - How to seamlessly combine those geometric approaches with dynamic controllers (ex. GIC)?

• Passive Velocity Field Control (PVFC)
  - PVFC defined on SE(3)
  - How to learn vector fields?

• Diffeomorphic transformation and their connection to the manifold-based approaches
  - Trajectory planning based on the manifold
  - Potential function learning on the manifold - look to the next bullet.

• Learning Potential Fields and resulting Velocity/Force field
  - In an equivariant manner
  - Potential field defined on the manifold

• Geometric Deep Learning
  - SE(3) Invariant/Equivariant Neural Networks for learning transferability and sample efficiency
  - Learning from Proprioception inputs - utilization of spherical harmonics

• Reinforcement Learning for Robotic Manipulation Tasks
  - End-to-end Deep Reinforcement Learning (DRL) is not my interest; my apologies.
  - How to formulate DRL into a more simple problem: Parameterization of tasks for DRL to easily deal with
  - Reinforcement learning as a soft constraint on policy

Previous Researches

• Nonlinear Control in application to UAVs
  - Approximate Dynamic Inversion & Extended High-gain Observers
  - Explicit Nonlinear Model Predictive Control
  - Recursive Least Square-based Dynamic Inversion

• Deep Reinforcement Learning of a semi-active suspension system

• Gain-scheduling Robust Control using LPV/LMI approach
  - Road-adaptive semi-active suspension system with LSTM-based road parameter estimator
  - Autonomous vehicle control under vision uncertainties

• Safety-critical control using Control Barrier Function
  - Adaptive cruise control for autonomous vehicles based on relaxed control barrier function and dynamic inversion

Contact me at joohwan_seo@berkeley.edu for Collaboration and discussions!