houchangmeng/safe-e2e-driving

End-to-end Autonomous Driving With Safety Constraints.

End-to-end Autonomous Driving With Safety Constraints

This is a PyTorch implementation of "End-to-end Autonomous Driving with Safety constraints" proposed in:

• "End-to-end Safe Autonomous Driving with Safety constraints" by Changmeng Hou, and Wei Zhang.

Requirements

Ubuntu 20.04
Carla 0.9.13
Python 3.7.12
torch==1.13.1

Installation

1. Set up conda environment.

$ conda create -n env_name python=3.7
$ conda activate env_name

2. Clone this git repo to an appropriate folder.

$ git clone git@github.com:houchangmeng/safe-e2e-antunomous.git

3. Enter the root folder of this repo and install the packages.

$ pip install -r requirements.txt

4. Train the safe agent .

$ python train_safe.py

Result

Results on custom carla reinforcement environment are as follows.

Ramp merging

Roundabout

Intersection

• The first row shows the ground truth of camera, lidar, and semantic birdeye images when the trained agent is running in the simulated town with surrounding vehicles and walkers (green boxes).
• The second row shows the reconstructed camera, lidar, and semantic birdeye images from the latent state, which is inferred online with the learned sequential latent model.

Sample a sequence

Sample a sequence of camera, lidar and birdeye images.

• The first row shows the ground truth.
• The secend row shows the posterior sample.
• The third row shows the conditional prior sample.
• The forth row shows the prior sample.
• Left to right indicates flowing time steps.