Hon Hai launches a deep learning model "QCNet" for autopilot trajectory prediction in the new era.

Thank you, CTOnews.com netizens. Hello, rabbit clue delivery! CTOnews.com May 23, the artificial Intelligence Institute of Hon Hai Research Institute of Foxconn Group has cooperated with City University of Hong Kong to launch a new generation of autopilot trajectory prediction deep learning model "QCNet", which can be applied to the group's electric vehicle autopilot system in the future.

CTOnews.com learned from the official website of Hon Hai that QCNet is an intelligent trajectory prediction model, and its main advantage is that it can understand the global information of the real driving scene, and modify the Transformer architecture with the same technical foundation of ChatGPT to be suitable for self-driving scene to learn the diversity and uncertainty of vehicle history trajectory, vehicle interaction behavior, road environment and so on.

It is reported that QCNet can predict the trajectory of the vehicle in the next 6 to 8 seconds, and accurately predict multiple targets in the scene, and the computational efficiency of the encoder is improved by 85%. At the same time, QCNet establishes a set of local coordinate system for each scene element in the traffic scene, such as driveway, zebra crossing, vehicle, pedestrian and so on, learns the representation in the local coordinate system, and captures the relative relationship between different scene elements through relative spatio-temporal position coding to improve the real-time computing efficiency of the model.