 |
| Source: NUS. Dr Ang posing with SMART-NUS autonomous vehicles. |
Can autonomous vehicles, popularly known as self-driving cars, replace conventionally-driven cars in the near future? It depends, says Dr Marcelo H Ang, Jr, Acting Director, Advanced Robotics Center and Associate Professor, Department of Mechanical Engineering, National University of Singapore (NUS), whose research remit includes working on autonomous vehicles.
Dr Ang is a pioneer in Singapore's autonomous vehicle scene. As far back as 2015 Dr Ang partnered with the Singapore-MIT Alliance for Research and Technology (SMART) on self-driving cars. The prototype, called SCOT (Shared Computer Operated Transport), began a year-long trial at the One-North business park in Singapore that year.
According to Dr Ang there are vehicles which move in a pedestrian environment, or where there are no roads, such as the mobility scooter and wheelchairs. There is last-mile urban driving, which brings passengers from their homes to the nearest train station. Another use case is the golf cart or golf buggy, meant for transportation inside large properties such as airport terminals or inside tourist attractions. “(There's) no fixed route. It just goes where you want, and is self-driving,” Dr Ang said.
Self-driving golf buggies, working in a constrained environment where traffic is light are already in use, Dr Ang said, and debuted in Singapore in 2014. Each vehicle drives up to a waiting passenger, transports the person to a destination, and then moves on to pick up the next passenger waiting elsewhere. The question is one of optimisation now, Dr Ang noted. “(When we) have more vehicles, it opens up an interesting question: how many vehicles are needed to provide mobility on demand, to guarantee that no one waits more than 5 minutes and gets to the destination in about 5 minutes?” he explained.
The next question is bridging the gap between academic research and commercialisation, Dr Ang said. “Every time we did a demonstration, people would ask how much it cost (to buy one),” he said. “My solution is asking if they have an engineering team I can work with.
“In universities, we stop at proofs of concept. We need to partner companies who see the value in this and more importantly returns on investment.”
ST Kinetics is one such company, and is currently a NUS partner for autonomous vehicles, he said.
Going 'live' on roads in a constrained environment with little traffic is the next step, and Dr Ang believes it could happen widely in five to seven years. Use cases would include fixed route services like a shuttle bus within the NUS compound, which goes along exactly the same route every day, or ferrying people who live in residential areas to nearby train stations.
“For free driving, it'll take at least 10 years. That's my best guess,” said Dr Ang.
Dr Ang is working on making cars smarter with the help of optimised technology such as the NVIDIA Drive PX 2 open artificial intelligence (AI) car computing platform. NUS has used NVIDIA solutions for autonomous vehicles for years, is familiar with the NVIDIA NVIDIA Jetson TK1 developer kit and For example, he explained that self-driving cars must be equipped with some way to determine location and the limitations of every point of the neighbourhood.
“If you are new to a place, for example if you are new to Singapore, I handhold you and say 'this is Singapore'. If you are a computer, you memorise; you see everything and you never forget. The next time (I leave you anywhere in Singapore) you know where you are. We need to do our own mapping. We have to navigate, we have to localise. That's the negative thing,” he said. “We are working on methods which do not require prior mapping.”
What Dr Ang is working towards is eliminating the need to use GPS as the signal can sometimes be unreliable, and enabling the vehicle to use natural features on roads to navigate instead, as humans do. For example, a human might just refer to a map, then draw on an understanding of road rules and nearby signage to determine how to get to a destination. Humans may not think much about walking or driving straight through an exit, but a vehicle has to be taught everything from first principles, Dr Ang pointed out, from recognising what an exit may look like to what to do to move through one, and doing it in a legal way.
“That's where a platform like NVIDIA is very helpful. It can differentiate landmarks. If you are in Singapore for the first time, I give you a map, and you know (where) you are and can find your way. Humans can do it but cars cannot do it. It's very interesting to use deep learning to do that,” he said. “After months of driving autonomously the car gets better and better at driving. That's why I'm very excited to use the NVIDIA platform for doing that.”
“Of all of the robotic applications, the robotic car (autonomous vehicle) is the most interesting. It has the most impact on everyday life. It's about giving mobility to people,” he said. “In a car the actuation (movement) is relatively easy - turning the steering wheel, stepping on the brakes, stepping on the accelerator, but perception, planning and prediction are complicated.”
According to Dr Ang there are vehicles which move in a pedestrian environment, or where there are no roads, such as the mobility scooter and wheelchairs. There is last-mile urban driving, which brings passengers from their homes to the nearest train station. Another use case is the golf cart or golf buggy, meant for transportation inside large properties such as airport terminals or inside tourist attractions. “(There's) no fixed route. It just goes where you want, and is self-driving,” Dr Ang said.
Self-driving golf buggies, working in a constrained environment where traffic is light are already in use, Dr Ang said, and debuted in Singapore in 2014. Each vehicle drives up to a waiting passenger, transports the person to a destination, and then moves on to pick up the next passenger waiting elsewhere. The question is one of optimisation now, Dr Ang noted. “(When we) have more vehicles, it opens up an interesting question: how many vehicles are needed to provide mobility on demand, to guarantee that no one waits more than 5 minutes and gets to the destination in about 5 minutes?” he explained.
The next question is bridging the gap between academic research and commercialisation, Dr Ang said. “Every time we did a demonstration, people would ask how much it cost (to buy one),” he said. “My solution is asking if they have an engineering team I can work with.
“In universities, we stop at proofs of concept. We need to partner companies who see the value in this and more importantly returns on investment.”
ST Kinetics is one such company, and is currently a NUS partner for autonomous vehicles, he said.
Going 'live' on roads in a constrained environment with little traffic is the next step, and Dr Ang believes it could happen widely in five to seven years. Use cases would include fixed route services like a shuttle bus within the NUS compound, which goes along exactly the same route every day, or ferrying people who live in residential areas to nearby train stations.
“For free driving, it'll take at least 10 years. That's my best guess,” said Dr Ang.
Dr Ang is working on making cars smarter with the help of optimised technology such as the NVIDIA Drive PX 2 open artificial intelligence (AI) car computing platform. NUS has used NVIDIA solutions for autonomous vehicles for years, is familiar with the NVIDIA NVIDIA Jetson TK1 developer kit and For example, he explained that self-driving cars must be equipped with some way to determine location and the limitations of every point of the neighbourhood.
“If you are new to a place, for example if you are new to Singapore, I handhold you and say 'this is Singapore'. If you are a computer, you memorise; you see everything and you never forget. The next time (I leave you anywhere in Singapore) you know where you are. We need to do our own mapping. We have to navigate, we have to localise. That's the negative thing,” he said. “We are working on methods which do not require prior mapping.”
What Dr Ang is working towards is eliminating the need to use GPS as the signal can sometimes be unreliable, and enabling the vehicle to use natural features on roads to navigate instead, as humans do. For example, a human might just refer to a map, then draw on an understanding of road rules and nearby signage to determine how to get to a destination. Humans may not think much about walking or driving straight through an exit, but a vehicle has to be taught everything from first principles, Dr Ang pointed out, from recognising what an exit may look like to what to do to move through one, and doing it in a legal way.
“That's where a platform like NVIDIA is very helpful. It can differentiate landmarks. If you are in Singapore for the first time, I give you a map, and you know (where) you are and can find your way. Humans can do it but cars cannot do it. It's very interesting to use deep learning to do that,” he said. “After months of driving autonomously the car gets better and better at driving. That's why I'm very excited to use the NVIDIA platform for doing that.”
“Of all of the robotic applications, the robotic car (autonomous vehicle) is the most interesting. It has the most impact on everyday life. It's about giving mobility to people,” he said. “In a car the actuation (movement) is relatively easy - turning the steering wheel, stepping on the brakes, stepping on the accelerator, but perception, planning and prediction are complicated.”
Interested?
Read about the five different levels of autonomous driving (PDF)
Watch Dr Ang speak on video about the SMART-NUS self driving car at the beginning of the trial at One-North
 |
| Source: NUS. The NUS team with the SMART-NUS cars. |
Watch Dr Ang speak on video about the SMART-NUS self driving car at the beginning of the trial at One-North
Watch a video introducing self-driving cars by NVIDIA from CES 2017. CES 2017 was in January 6 to 9, 2017. More advanced technology can be expected at CES 2018, from January 9 to 12, 2018
No comments:
Post a Comment