Use first 10 days of ATC dataset, first day for train and the rest for evaluation. The data is in atc_data/middle_area folder, and add to git lfs.
Use middle area of ATC dataset. This middle area has dimensions where x ranges from -25m to 0m and y ranges from -10m to 15m, amounting to an area of 625m2, which is shown in the figure below.
Use CLiFF-LHMP and Trajectron++ as baselines. The config file of training Trajectron++ is attached in /baselines/trajectron++_config.json
Main entrance of LaCE-LHMP is in main_lace_lhmp.py. Can run
python3 main_lace_lhmp.py --r_s 1 --beta 1 --generate_traj_num 5 --version "version1" --observed_tracklet_length 3 --planning_horizon 20 --delta_t 1
or add --no_rank to evaluate without ranking, but compute average ADE/FDE of k generated trajs.
One example of running result is:
ADE/FDE value in results/version1/1031.csv
prediction figure in quality_res/LT_9103800_in_date1031.png
