Inria Industry Meetings
Computer Vision and Dynamic Planning for Autonomous Vehicles
Demonstration of two research projects by the RITS team in conjunction with Valeo.
- 1) While convolutional neural networks (CNNs) have been designed to process dense data, scattered data is inevitable in vision. We offer robust architecture for high density input data. Performance is demonstrated on two crucial vision tasks: completion and pixel-wise semantic annotation.
- 2) The Reachable Interaction Sets (RIS) presented here use a dynamic representation of the environment that anticipates the movements of obstacles and of the robot. The time dimension is no longer needed on a new 2-dimensional map, which means that classic path-finding methods can be used.