LSTM_EGO

Learn to navigate in dynamic environments with normalized LiDAR scans, video

Configuration

• DRL structure is based on the open-source solution from https://github.com/AntoineTheb/RNN-RL
• Pytorch
• Cython
• RVO2
• socialforce
• ubuntu 20.04, ros-noetic

Compile LiDAR scan library

In C_library, python setup.py build_ext --inplace

Choose environment

In main.py, parser.add_argument("--env", default="crowd_real_all_circles"). "crowd_sim" stands for basic environments only having 5 dynamic circle humans; "crowd_sim_complex" represent complex scenarios mixing dynamic circle humans (maximum number is 4) and static rectangle obstacles (maximum number is 3); "crowd_real" denotes real-world environments where the robot moves on a narrow plane; "crowd_real_all_circles" are the environments for sim-to-real transfer.

Compile SLAM and Clustering algorithms

• codes are based on the open-source solution from https://github.com/HKUST-Aerial-Robotics/A-LOAM and https://github.com/yzrobot/adaptive_clustering
• mkdir ws_slam_cluster
• cd ws_slam_cluster
• mkdir src
• copy SLAM_Cluster into ws_slam_cluster/src
• catkin_make
• source devel/setup.bash
• roslaunch aloam_velodyne adaptive_clustering.launch
• please define the detection area in adaptive_clustering.cpp, line 120

drl_vo is a supplementary baseline, please refer to this link