LiDAR-as-Camera for End-to-End Driving

TL;DR

This paper demonstrates that LiDAR-images from Ouster sensors can effectively replace cameras for end-to-end autonomous driving, providing aligned, multi-channel data that performs comparably in real-world road-following tasks and offers insights into prediction smoothness and driving ability.

Contribution

It introduces the use of LiDAR-images for end-to-end driving, showing they are sufficient and align well with visual data, and analyzes the correlation between prediction smoothness and driving performance.

Findings

LiDAR-images perform at least as well as camera-based models in road-following.

LiDAR data simplifies sensor alignment issues in real-world driving.

Temporal smoothness of predictions correlates with driving ability.

Abstract

The core task of any autonomous driving system is to transform sensory inputs into driving commands. In end-to-end driving, this is achieved via a neural network, with one or multiple cameras as the most commonly used input and low-level driving command, e.g. steering angle, as output. However, depth-sensing has been shown in simulation to make the end-to-end driving task easier. On a real car, combining depth and visual information can be challenging, due to the difficulty of obtaining good spatial and temporal alignment of the sensors. To alleviate alignment problems, Ouster LiDARs can output surround-view LiDAR-images with depth, intensity, and ambient radiation channels. These measurements originate from the same sensor, rendering them perfectly aligned in time and space. We demonstrate that such LiDAR-images are sufficient for the real-car road-following task and perform at least equally to camera-based models in the tested conditions, with the difference increasing when needing to generalize to new weather conditions. In the second direction of study, we reveal that the temporal smoothness of off-policy prediction sequences correlates equally well with actual on-policy driving ability as the commonly used mean absolute error.