Accelerating autonomous vehicle research at the National University of Singapore with NVIDIA

Source: NVIDIA, NUS. The NUS team with SMART-NUS autonomous vehicles. The Singapore-MIT Alliance for Research and Technology (SMART) is a research enterprise established by the Massachusetts Institute of Technology (MIT) in partnership with the National Research Foundation of Singapore (NRF) in 2007.

The NVIDIA DRIVE PX 2 AI car computing platform is helping researchers implement key artificial intelligence (AI) drive algorithms for autonomous vehicle insight.

The challenge is how self-driving cars can be made even safer on the roads. Self-driving cars already know how to navigate safely along roads. However, the university’s research is now focused on using deep learning to predict the behaviour of objects around the vehicle. These objects vary from traffic signals to pedestrians. Because training was taking a long time, and required massive amounts of compute power for processing, the National University of Singapore (NUS) was looking for a specialised solution that could solve its autonomous vehicle (AV) R&D challenges.

The solution? NVIDIA Tesla GPUs and the DRIVE PX 2 open AI car computing platform are shortening the R&D time for the existing AV research at NUS. Two units of NVIDIA DRIVE PX 2 systems and the accompanying software development kit (SDK) are helping the research team to test AI algorithm inferencing. For the training side, multiple Pascal graphics processing units* (GPUs) are used to power the neural networks used for object recognition semantics. This technology suite powers the university’s experimental AVs as well assists partner Singapore Technologies Kinetics on their autonomous bus project.

The DRIVE PX 2 is optimised for AV research and handles sensor fusion using complex algorithms, facilitating NUS researchers to focus on making AVs more intelligent at a faster rate. They aim to achieve better self-driving performance using deep learning, which mimics the synapses of the human brain. Also, NUS is conducting accurate, in-depth research on AVs at a lower cost per hour using NVIDIA’s Pascal GPUs and DRIVE PX 2 platform for training and inferencing as compared to using other high-performance computing solutions. Additionally, NVIDIA Pascal GPUs complete training sessions in significantly shorter times, keeping NUS at the forefront of global AV research.

Source: NUS, NVIDIA. Dr Ang standing among SMART-NUS autonomous vehicles.

“With self-driving cars, one of the hardest things to predict is how the vehicle would behave and predict like a human.. With NVIDIA DRIVE PX 2 we don't need to reinvent the wheel; we can build on top of the best of what is already available, and in turn letting us concentrate more on building new capabilities,” said Dr Marcelo H Ang, Jr, Acting Director, Advanced Robotics Center and Associate Professor, Department of Mechanical Engineering, National University of Singapore.

Interested?

Read the TechTrade Asia blog posts about how NVIDIA is creating an AI-based ecosystem for autonomous vehicles and Dr Ang's take on where autonomous vehicles need to go next in Singapore

*GPUs are so named as opposed to CPUs, central processing units. Traditionally a CPU controlled the computer, and a GPU would take care of how graphics is handled on the device. Today complex computing requirements are such that a GPU can take over control as effectively or even more effectively than a CPU.