Conditional Latent ODEs for Motion Prediction in Autonomous Driving

TL;DR

This paper introduces cLODE, a novel model combining conditional VAE and neural ODE for improved multi-agent motion prediction in autonomous driving, demonstrating superior performance and efficiency over existing methods.

Contribution

The paper proposes the conditional latent ODE (cLODE), integrating generative and continuous modeling techniques for enhanced motion prediction in autonomous driving scenarios.

Findings

Outperforms baseline methods in multi-agent driving simulation

Uses less GPU memory compared to previous models

Provides publicly available code and Docker image

Abstract

This paper addresses imitation learning for motion prediction problem in autonomous driving, especially in multi-agent setting. Different from previous methods based on GAN, we present the conditional latent ordinary differential equation (cLODE) to leverage both the generative strength of conditional VAE and the continuous representation of neural ODE. Our network architecture is inspired from the Latent-ODE model. The experiment shows that our method outperform the baseline methods in the simulation of multi-agent driving and is very efficient in term of GPU memory consumption. Our code and docker image are publicly available: https://github.com/TruongKhang/cLODE; https://hub.docker.com/r/kim4375731/clode.