Sampling-based motion planning(SMPs) approach has been very popular for its ability of computing collision-free and asymptotically optimal path without explicit formulation of the configuration space. SMPs use sampling to generate a discrete representation of the problem and then run graph searching algorithm on this representation. Which means the representation itself is at least as important as graph searching algorithm. In general this is enabled by uniformly sampling the configuration space. This paper proposes a machine learning based biased sampling approach for autonomous driving. The sampling distribution was learned from previous demonstrations using conditional variational encoder(CVAE) with attention mechanism. Combined with a sampling-based algorithm called rapidly-exploring random tree∗(RRT∗), we proposed Attention-RRT∗. This approach was proved to be effective in several driving scenarios.

Attention-based Sampling Distribution for Motion Planning in Autonomous Driving

Arrigoni S.;Braghin F.
2020

Abstract

Sampling-based motion planning(SMPs) approach has been very popular for its ability of computing collision-free and asymptotically optimal path without explicit formulation of the configuration space. SMPs use sampling to generate a discrete representation of the problem and then run graph searching algorithm on this representation. Which means the representation itself is at least as important as graph searching algorithm. In general this is enabled by uniformly sampling the configuration space. This paper proposes a machine learning based biased sampling approach for autonomous driving. The sampling distribution was learned from previous demonstrations using conditional variational encoder(CVAE) with attention mechanism. Combined with a sampling-based algorithm called rapidly-exploring random tree∗(RRT∗), we proposed Attention-RRT∗. This approach was proved to be effective in several driving scenarios.
PROCEEDINGS OF THE 39TH CHINESE CONTROL CONFERENCE
978-9-8815-6390-3
Autonomous Driving
Machine Learning
Motion Planning
File in questo prodotto:
File Dimensione Formato  
09189020.pdf

Accesso riservato

Descrizione: paper
: Publisher’s version
Dimensione 510.45 kB
Formato Adobe PDF
510.45 kB Adobe PDF   Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11311/1163441
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 1
  • ???jsp.display-item.citation.isi??? 0
social impact