Foreground IP for licensing: 1. Positioning & Navigation Core a. Map modeling and management i. implement at least 2 map models out of general topological, geometric, hybrid topological, semantic, etc. b. Marker-based positioning i. handle at least 2 types of markers, such as QR-code, barcode, frame marker, etc. ii. simultaneous, multiple barcode recognition based on captured video c. Marker-based navigation 2. Setup of centralized computing platform for deep learning 3. Route planning based on Reinforcement deep learning a. Single device route learning strategies b. Multiple device route learning strategies c. Adaptive routing implementation empowered by deep reinforcement learning Please refer to the document "Technical_Supplement.pdf" for specifications of the deliverables.
Sharp Asia Computer Services Ltd.
As the autonomous vehicle industry racing from zero to warp speed, it calls for technology innovations in every aspect of this field. For automated guided vehicle (AGV), one type of autonomous vehicles, its global market alone is projected to be US$2.68B by 2022. In an environment of hundreds or even thousands of vehicles, the technical challenges include looking for the most efficient routes for all vehicles and the optimal assignment of vehicles to perform their tasks, the whole scalability issue, avoiding obstacles, etc. In this project, ASTRI will develop a platform to enable the scheduling and operations of autonomous devices, with deep learning incorporated to improve on dynamic route optimization, large-scale vehicle planning and task scheduling, as well as visual-based collision avoidance. ASTRI's initial applications will be to AGV for distribution centers as well as to training of autonomous vehicle control.