Package architecture

[Geonhee-LEE]

Package architecture:

• controllers:

  class control: PID, pure pursuit, bang-bang, open-loop(velocity profile), ... optimal control: lqr, ddp, mpc, ... collision avoidance: RVO, ORCA, ...

• policies:

  (Model-free)

  • value based RL: DQN, ...,
  • policy based RL: DDPG, TRPO, SAC, ...
  • actor critic RL: SAC, ... Imitation learning, Inverse RL Model-based RL: MBPO, MB-TRPO, GPS, ... paper implementation(CADRL, MACA, ...)

• envs:

  When using controllers, it support the start and goal pose, trajectory from model Otherwise, support openAI gym style (observe, step, action, train)

• sensors:

  2D Lidar camera OccupancyGridSensor People detection

• models:

  kinematic, dynamics For model-based RL, NN model, GP model, LSTM model

• utils:

  Every utilities such as visualization, calculation, optimization

• configs:

  saved policy expert demonstration configuration parameters map data,

References:

• https://github.com/liuzuxin/MPC_template-model_predictive_control_for_reinforcement_learning
• https://github.com/ray-project/ray
• https://github.com/rail-berkeley/rlkit
• https://github.com/mit-acl/gym-collision-avoidance/blob/release/gym_collision_avoidance/envs/sensors/OccupancyGridSensor.py
• https://github.com/Geonhee-LEE/rl_local_planner

[Geonhee-LEE]

https://github.com/Shunichi09/PythonLinearNonlinearControl/tree/master/PythonLinearNonlinearControl/models 을 크게 참조하여 구성