Proposal Discussions

  1. ✔️Principal RL / control theoretical RL / Structured RL / NN symbolic RL

    1. Integrate Control Theory into RL. Safe Guarantee.

    2. Possible task: UAVs high-speed swarm control; heterogeneous multi-agent system;

    3. Resources:

      End-to-End Safe Reinforcement Learning through Barrier Functions for Safety-Critical Continuous Control Tasks

  2. sim2real

  3. multi-modality

Project Comments

  1. formation control project
    1. Lack of intellectual novelty: a multi-goal RL project (actually have a global observation.) → theory guarantee
    2. Safe guarantee: no collision in real world. → RL for high-level control + safe constrains.
  2. bimanual manipulation
    1. better way to exploit symmetry: Geometric Deep Learning

TODO

  1. Take course
  2. Literature review & baselines
  3. Hardware system setup

Resources

  1. courses

    Control Bootcamp

  2. papers

  3. open-source code

    Welcome to Crazyswarm's documentation! - Crazyswarm 0.3 documentation