Researchers have combined their expertise to create RealMotion, a novel training system that seamlessly integrates historical and real-time scene data with contextual and time-based information, paving the way for more efficient and reliable autonomous vehicle technology. The related research paper has been published on the arXiv preprint server.
Dr. Xiatian Zhu, a senior lecturer at the Centre for Vision, Speech and Signal Processing and the Human-Centered AI Institute at the University of Surrey and a co-author of the study, said: “Driverless cars are no longer a dream of the future. Robotaxis are already in operation in parts of the US and China, and autonomous vehicles are expected to be on UK roads as early as next year. However, the real question in everyone’s mind is: how safe are they? While artificial intelligence works differently from human drivers, there are still challenges to overcome. That’s why we developed RealMotion – not only equipping the algorithm with real-time data, but also enabling it to integrate spatiotemporal historical context, thus making more accurate and reliable decisions and achieving safer autonomous navigation.”
Existing motion prediction methods typically process each driving scenario independently, ignoring the interrelationships between past and present environments in continuous driving scenarios. This limitation hinders the ability to accurately predict the behavior of surrounding vehicles, pedestrians, and other agents in a constantly changing environment.
In contrast, RealMotion can gain a clearer understanding of different driving scenarios. Integrating past and present data enhances the prediction of future motion, addressing the inherent complexity of predicting multiple agent motions.
Extensive experiments using the Argoverse dataset, a leading benchmark in autonomous driving research, highlight the accuracy and performance of RealMotion. Compared to other AI models, the framework achieves an 8.60% improvement in Final Displacement Error (FDE) (the distance between the predicted final destination and the actual final destination). It also demonstrates a significant reduction in computational latency, making it well-suited for real-time applications.
Professor Adrian Hilton, Director of the Human-Centered AI Institute in Surrey, said: “With autonomous vehicles soon to be on the roads in the UK, ensuring people’s safety is of paramount importance. Dr. Zhu and his team’s RealMotion represents a significant advancement over existing methods. By enabling autonomous vehicles to perceive their surroundings in real time and make informed decisions using historical context, RealMotion paves the way for safer and smarter navigation on our roads.”
Despite some limitations encountered by the researchers, the team plans to continue their work to further improve RealMotion's capabilities and overcome various challenges. This framework has the potential to play a crucial role in shaping the next generation of autonomous vehicles, ensuring that future navigation systems are safer and smarter.
