eduzhai > Applied Sciences > Engineering >

Sample Efficient Interactive End-to-End Deep Learning for Self-Driving Cars with Selective Multi-Class Safe Dataset Aggregation

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 6 pages

Abstract: The objective of this paper is to develop a sample efficient end-to-end deeplearning method for self-driving cars, where we attempt to increase the valueof the information extracted from samples, through careful analysis obtainedfrom each call to expert driverś policy. End-to-end imitation learning is apopular method for computing self-driving car policies. The standard approachrelies on collecting pairs of inputs (camera images) and outputs (steeringangle, etc.) from an expert policy and fitting a deep neural network to thisdata to learn the driving policy. Although this approach had some successfuldemonstrations in the past, learning a good policy might require a lot ofsamples from the expert driver, which might be resource-consuming. In thiswork, we develop a novel framework based on the Safe Dateset Aggregation (safeDAgger) approach, where the current learned policy is automatically segmentedinto different trajectory classes, and the algorithm identifies trajectorysegments or classes with the weak performance at each step. Once the trajectorysegments with weak performance identified, the sampling algorithm focuses oncalling the expert policy only on these segments, which improves theconvergence rate. The presented simulation results show that the proposedapproach can yield significantly better performance compared to the standardSafe DAgger algorithm while using the same amount of samples from the expert.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×