Efficient Latent Representations using Multiple Tasks for Autonomous Driving

TL;DR

This paper introduces a multi-task encoder-decoder neural network to learn low-dimensional, informative environment representations for autonomous driving, improving learning speed and performance in data-scarce scenarios.

Contribution

It presents a novel multi-head neural network architecture for environment representation, outperforming single-head models in autonomous driving tasks.

Findings

Faster policy learning with the proposed multi-task representation.

Enhanced performance in autonomous driving scenarios.

Reduced data requirements for effective policy training.

Abstract

Driving in the dynamic, multi-agent, and complex urban environment is a difficult task requiring a complex decision policy. The learning of such a policy requires a state representation that can encode the entire environment. Mid-level representations that encode a vehicle's environment as images have become a popular choice, but they are quite high-dimensional, which limits their use in data-scarce cases such as reinforcement learning. In this article, we propose to learn a low dimensional and rich feature representation of the environment by training an encoder-decoder deep neural network to predict multiple application relevant factors such as trajectories of other agents. We demonstrate that the use of the multi-head encoder-decoder neural network results in a more informative representation compared to a single-head encoder-decoder model. In particular, the proposed representation learning approach helps the policy network to learn faster, with increased performance and with less data, compared to existing approaches using a single-head network.