Presentation for explaining the paper "DTPP: Differentiable Joint Conditional Prediction and Cost Evaluationfor Tree Policy Planning in Autonomous Driving" presented from Nanyang Technological University, NVIDIA and Stanford University.
In recent years, DNN models have achieved fairly high recognition accuracy in camera images for autonomous driving. However, downstream vehicle control seems to be a major issue.
This paper proposes a method to estimate the states of ego vehicle and nearby vehicles in a tree structure, and use Transformer to find optimal trajectory planning from multiple possible future states of ego and other vehicles.