ZhaiMen-Hub/DecisionPlanning

The repository implements some decision-making and contingency planning (contingency model predictive control) methods.

DecisionPlanning

The repository implements some decision-making and contingency planning methods.

If you find this work useful or interesting, please kindly give me a star ⭐, thanks!😀

Requirements

• System: Ubuntu 20.04 (Tested on Win11 WSL2 )
• Python 3.8.10: numpy == 1.24.4 matplotlib == 3.7.4 casadi == 3.6.5

Contingency MPC / Contingency Planning

How to run

• Navigate to the root folder

• Uncomment the corresponding code block of settings in ContingencyMPC/CMPC.py (enable CMPC - obstacle pops out as follows)

  # CMPC - obstacle pops out
  obs_pop = True
  Pc = 0.25        # Cost weight for contingency control
  
  # # CMPC - obstacle does not pop out
  # obs_pop = False
  # Pc = 1e-2       # for unique minimum solution
  
  # # RMPC
  # obs_pop = False
  # Pc = 1.0 - 1e-2   

• Run the script python3 ContingencyMPC/CMPC.py

• The anime and figures are saved in ContingencyMPC/log/

Results

• CMPC - obstacle pops out (at k = 4)
  • at k = 5, the nominal planner observes this movement and recognizes the contingency has occurred

• CMPC - obstacle pops out (at k=4)

• RMPC
  • assumes the worst case evolution of the scene (follow the contingency planner, the red triangles), even though the obstacle never pops out

Reference

• Alsterda, John P. and J. Christian Gerdes. “Contingency Model Predictive Control for Linear Time-Varying Systems.” ArXiv abs/2102.12045 (2021): n. pag.
  • Contingency Model Predictive Control for Linear Time-Varying Systems - 知乎 (zhihu.com)
• Alsterda, John P., Matthew Brown and J. Christian Gerdes. “Contingency Model Predictive Control for Automated Vehicles.” 2019 American Control Conference (ACC) (2019): 717-722.

Game-Theoretic Motion Planning

How to run

• Uncomment the corresponding code block of settings in Game-Theoretic Motion Planning/Game.py

  KF = 0.01
  KL = 1 - KF
  # distF = 20    # collision ditance (conservative)
  distF = 10    # collision ditance (agressive)
  distL = 15
  Kinfluence = 0
  
  # KF = 0.5
  # KL = 1 - KF
  # distF = 20    # collision ditance
  # distL = 20
  # Kinfluence = 1 # add Jinfluence

• Run the script python3 Game-Theoretic Motion Planning/Game.py

• The anime and figures are saved in Game-Theoretic Motion Planning/log/

Results

• Aggressive follower: Leader (automated vehicles) performs an aggressive lane change

• Conservative follower: Leader accelerates before performing lane change (also the Fig 4. of [Burger, 2022])

Reference

• Burger, Christoph, Johannes Fischer, Frank Bieder, Ömer Sahin Tas and Christoph Stiller. “Interaction-Aware Game-Theoretic Motion Planning for Automated Vehicles using Bi-level Optimization.” 2022 IEEE 25th International Conference on Intelligent Transportation Systems (ITSC) (2022): 3978-3985.
  • Interaction-Aware Game-Theoretic Motion Planning for Automated Vehicles using Bi-level Optimization - 知乎 (zhihu.com)
  • 论文复现 基于博弈论的运动规划 - 知乎 (zhihu.com)
• Burger, Christoph and Martin Lauer. “Cooperative Multiple Vehicle Trajectory Planning using MIQP.” 2018 21st International Conference on Intelligent Transportation Systems (ITSC) (2018): 602-607.

Game-Theoretic CMPC

How to run

• Uncomment the corresponding code block of settings in Game-Theoretic Motion Planning/CMPCGame.py

  # scenario
  
  decel = True        # whether follower will decelarate
  # decel = False

• Run the script python3 Game-Theoretic Motion Planning/CMPCGame.py

• The anime and figures are saved in Game-Theoretic Motion Planning/log/

Results

• Follower decelerate

• Follower does not decelerate

Reference

• Wei, Junqing, John M. Dolan and Bakhtiar Litkouhi. “Autonomous vehicle social behavior for highway entrance ramp management.” 2013 IEEE Intelligent Vehicles Symposium (IV) (2013): 201-207.
  • Autonomous Vehicle Social Behavior for Highway Entrance Ramp Management - 知乎 (zhihu.com)
• [附代码]融合Leader-Follower Game和Contingency MPC的预测决策规划 - 知乎 (zhihu.com)