Planning-oriented Autonomous Driving

https://arxiv.org/abs/2212.10156

Planning-oriented Autonomous Driving

---

 

Introduction

이 논문은 자율주행의 e2e framework를 소개하고 있으며, 각 task 별 모듈들을 하나의 네트워크로 통합한 full-stack explicit design을 채택했다. (2023 CVPR Best paper)

Contribution은 다음과 같다.

• 광범위한 task들을 하나로 통합한 planning-oriented e2e AD framework를 제안했고, 효과적인 task coordination의 필요성을 시사함.
• Unified query design을 제안하여 task간 연결을 용이하게 함.
• Realistic한 시나리오들에 대해 성능 검증을 진행했으며, vison-only input으로 SOTA 달성함.

 

Methodology

Figure 2. Pipeline of Unified Autonomous Driving (UniAD). It is exquisitely devised following planning-oriented philosophy. Instead of a simple stack of tasks, we investigate the effect of each module in perception and prediction, leveraging the benefits of joint optimization from preceding nodes to final planning in the driving scene. All perception and prediction modules are designed in a transformer decoder structure, with task queries as interfaces connecting each node. A simple attention-based planner is in the end to predict future waypoints of the ego-vehicle considering the knowledge extracted from preceding nodes. The map over occupancy is for visual purpose only.

Figure 2는 본 논문에서 제안한 UniAD의 전체 구조이다. 구체적인 논문 내용은 pdf로 첨부했다.

Planning_oriented_AD_review.pdf

2.59MB