Bird's-Eye-View Based Trajectory Planning of Multiple Mobil Robots Using DRL and MPC

Trajectory Planning of Multiple Robots using Vision-Based Continuous Deep Reinforcement Learning and Model Predictive Control

The main branch of this repository is the single robot implementation of our paper. The multiple robot implementation is in a separate branch in the repository.

The paper can be found here: Paper

Quick Start

OpEn

The NMPC formulation is solved using open source implementation of PANOC, namely OpEn. Follow the installation instructions before proceeding.

Install dependencies (after installing OpEn)

pip install -r requirements.txt

or

conda env create -f environment.yaml

NOTE If you cannot create the virtual environment via conda, please create your own virtual environment (e.g. conda create -n rlboost python=3.9), and pip install. Make sure your RUST is up-to-date and Pytorch is compatible with Cuda.

Generate MPC solver

Go to "test_block_mpc.py", change INIT_BUILD to true and run

python test_block_mpc.py

After this, a new directory mpc_build will appear and contain the solver. Then, you are good to go :)

To train the DDPG

Go to "src/continous_training_local.py", change load_checkpoint to False and run.

Use Case

Run src/main_continous.py for the simulation in Python. Several cases are available by changing scene_option in src/main_continous.py.

Extended result table

Scene	Obstacle Type	Method	Comp. time (mean)	Comp. time (max)	Comp. time (std)	Deviation (mean)	Deviation (max)	Speed	Angular speed	Finish time step	Success rate (%)
Scene 1	(a) Rectangular obstacle	MPC	29.03	94.11	24.51	0.54	1.94	0.04	0.03	76	100
Scene 1	(a) Rectangular obstacle	DDPG	0.77	10.41	7.78	1.02	1.98	0.08	0.61	73	100
Scene 1	(a) Rectangular obstacle	HYB-DQN-V	23.07	240.46	45.22	0.91	2.11	0.03	0.1	81	100
Scene 1	(a) Rectangular obstacle	HYB-DDPG	11.94	49.74	8.54	0.9	2.23	0.02	0.06	76	100
Scene 1	(b) Two obstacles	HYB-DQN-V	20.18	109.7	18.39	1.0	2.68	0.04	0.06	98	100
Scene 1	(b) Two obstacles	HYB-DDPG	17.38	86.21	14.31	0.96	2.62	0.03	0.05	97	100
Scene 1	(c) U-shape obstacle	DDPG	0.79	10.65	7.85	0.98	1.91	0.08	0.58	74	100
Scene 1	(c) U-shape obstacle	HYB-DQN-V	93.49	1019.9	254.09	0.84	2.93	0.09	0.16	118	40
Scene 1	(c) U-shape obstacle	HYB-DDPG	15.74	99.04	17.95	0.87	2.18	0.02	0.07	75	100
Scene 1	(d) Dynamic obstacle (face-to-face)	MPC	16.07	129.18	25.02	0.44	1.69	0.03	0.04	69	100
Scene 1	(d) Dynamic obstacle (face-to-face)	DDPG	0.82	10.96	7.95	0.74	1.94	0.08	0.48	77	98
Scene 1	(d) Dynamic obstacle (face-to-face)	HYB-DDPG	11.13	69.7	13.01	0.9	2.85	0.02	0.05	79	100
Scene 2	(e) Sharp turn with an obstacle	DQN-V	0.69	10.74	5.49	0.87	1.44	0.19	0.7	150	78
Scene 2	(e) Sharp turn with an obstacle	DDPG	0.74	10.61	5.35	0.53	1.37	0.09	0.56	158	96
Scene 2	(e) Sharp turn with an obstacle	HYB-DQN-V	11.8	135.32	15.95	0.49	1.8	0.02	0.04	149	100
Scene 2	(e) Sharp turn with an obstacle	HYB-DDPG	14.82	98.26	15.47	0.43	1.63	0.01	0.03	151	100
Scene 2	(f) U-turn with an obstacle	DQN-V	0.75	10.48	7.80	0.42	0.87	0.17	0.67	149	40
Scene 2	(f) U-turn with an obstacle	DDPG	0.73	10.48	5.00	0.63	1.05	0.13	0.51	169	74
Scene 2	(f) U-turn with an obstacle	HYB-DQN-V	14.19	99.66	15.10	0.43	1.67	0.02	0.03	148	100
Scene 2	(f) U-turn with an obstacle	HYB-DDPG	11.91	75.17	11.80	0.4	1.42	0.02	0.02	147	100

Evaluation videos

Videos of the evaluations are available on youtube

Navigation-of-Mobile-Robots-via-MPC-DRL

Navigation-of-Mobile-Robots-via-MPC-DRL

Navigation-of-Mobile-Robots-via-MPC-DRL