Precisive real-time prediction of the movement of nearby vehicles or the future trajectory of pedestrians is essential for safe autonomous driving. A research team from the City University of Hong Kong (CityU) has recently developed an innovative AI system that significantly improves predictive accuracy in dense traffic scenarios. This breakthrough technology also offers increased computational efficiency by over 85%, showing great potential for enhancing the safety of autonomous vehicles.
Existing solutions for behavior prediction in autonomous driving often face significant challenges in correctly understanding driving scenarios and lack efficiency in their predictions. These solutions typically involve re-normalizing and re-encoding the latest positional data of surrounding objects and the environment whenever the vehicle and its observation window move forward. However, this redundant computation and latency in real-time online predictions can lead to delays and errors, potentially causing catastrophic accidents.
In order to overcome the limitations of existing prediction models, Professor Wang Jianping and her team at CityU have introduced a revolutionary trajectory prediction model called “QCNet.” This model is based on the principle of relative space-time for positioning, offering unique and fixed position information regardless of the viewer’s space-time coordinate system. The QCNet model incorporates the relative positions of road users, lanes, and crosswalks, allowing it to capture their relationships and interactions in driving scenarios.
The QCNet model demonstrates two significant properties: “roto-translation invariance in the space dimension” and “translation invariance in the time dimension.” These properties enable the caching and reusing of previously computed encodings of coordinates, resulting in theoretically real-time operation. By leveraging these properties, QCNet can generate collision-free predictions while accounting for uncertainty in the future behavior of road users. This enhanced understanding of the rules of the road and interactions among multiple road users contributes to safer and more accurate predictions.
Evaluation of QCNet
To evaluate the efficacy of QCNet, the research team utilized two large-scale collections of open-source autonomous driving data and high-definition maps: “Argoverse 1” and “Argoverse 2.” These datasets are considered the most challenging benchmarks for behavior prediction and comprise over 320,000 sequences of data and 250,000 scenarios. In testing, QCNet demonstrated both speed and accuracy in predicting road users’ future movements, even with long-term predictions of up to six seconds. It outperformed 333 prediction approaches on Argoverse 1 and 44 approaches on Argoverse 2. Notably, QCNet significantly reduced online inference latency from 8ms to 1ms and increased efficiency by over 85% in dense traffic scenes involving 190 road users and 169 map polygons.
Integrating QCNet into autonomous driving systems can effectively enhance vehicles’ understanding of their surroundings and improve the accuracy of predicting the future behavior of other road users. This, in turn, enables safer and more human-like decision-making in autonomous vehicles. Professor Wang envisions applying this technology to various other applications in autonomous driving, including traffic simulations and human-like decision-making processes.
The research team from CityU has made significant strides in revolutionizing the prediction accuracy and computational efficiency of autonomous driving systems. Their QCNet model offers a breakthrough trajectory prediction solution that overcomes the limitations of existing models. With its ability to understand driving scenarios more accurately and efficiently, QCNet paves the way for safer autonomous driving. The team’s research findings have been presented at the prestigious IEEE / CVF Computer Vision and Pattern Recognition Conference and will be integrated into Hon Hai Technology Group’s autonomous driving system to enhance real-time prediction efficiency and self-driving safety.
Leave a Reply