Meet the researcher: autonomous vehicles, navigation systems and AI

Professor Michael Milford researches with QUT’s Science and Engineering Faculty, and is chief investigator at the Australian Centre for Robotic Vision.

When we, as humans, navigate between two locations, we’re making use of so many inherent and subconscious brain processes, it’s kind of spooky when you stop to think about it. We can look at our environment, see the hazards, identify the safe shortcuts, have an innate sense of direction relative to our position, know when to start and stop within the flux of traffic, and have a sense of self-preservation that keeps us on track.

Now, imagine teaching robots to do this.

Creating artificial intelligence (AI) that can navigate within unfamiliar environments is a huge challenge — just ask QUT’s Professor Michael Milford. He and his team have been developing technology for autonomous vehicles for years now. Drawing on inspiration from the natural world, they’ve developed navigation systems that can increase efficiency and safety across a range of industries.

What is an autonomous vehicle?

An autonomous vehicle (AV) is supported by an AI system that can learn behaviours and tactics to navigate an environment without human input. “That could be robots, or drones — basically anything that moves, that needs to know where it is and how to get from A to B,” Milford explained.

The hottest topic in AVs today is self-driving cars. “There’s a lot of hardware on the vehicle that you wouldn’t have on a normal car: sensors, cameras, LIDAR sensors, GPS. And then there’s the brains of the car, which is typically processing sensory input coming in from all those different sources and trying to work out how to drive.”

While referring to the AV’s computing system as its ‘brains’ might be anthropomorphising the technology, well, that’s kind of the point. “Basically, all the capabilities that humans do so well naturally, we’re trying to do in software,” Milford said.

“It’s important to remember that since the 1990s we’ve had autonomous vehicles like Navlab that could drive pretty well on the highway. The challenge is getting them to drive well in urban situations alongside pedestrians and cyclists. You can’t deploy a system that isn’t nearly perfect.

“The industry has realised that it’s a very tough problem. Some incidents where vehicles underperformed has prompted a lot of scrutiny on the industry — and rightfully so. I think that’s put us in a much healthier place in our attitudes toward the technology compared to five years ago.”

Taking a road trip

Milford has recently been testing the success of AVs on Queensland’s roads. Working with Queensland Department of Transport and Main Roads and the iMOVE Cooperative Research Centre, he and the QUT Research Engineering Facility team put together a car kitted out with all the sensors you’d find on a self-driving car, and took it on a 1,200km road trip around South-East Queensland.

Milford and team with their test vehicle, which completed a 1,200km road trip to test its accuracy. Image credit: QUT Media

They tested the system’s accuracy on things like sign detection, traffic light behaviour, interpretation of lane markings and internal positioning systems. They drove the car and compared the AI’s decisions to their expert human driving.

“It didn’t work that well,” Milford said. “But we found that by integrating a high definition map into the system’s ‘computer brain’, it could use what it already knew about the environment to make its performance so much better.”

‘Computer brain’ is, he maintains, a very technical term.

The exercise demonstrated that there’s still a lot of room to develop and grow within AV technology. The data they captured will help refine the functionality of their AI and will inform the types of systems established to manage AVs on our roads.

Mining the issue

A lot of Milford’s research and development focuses on solving problems in Australian context. As a result, he’s led a team developing AI solutions for the mining industry. Australia is a world leader in autonomous vehicle technology, with a focus on how technology can make the industry safer and more efficient.

“We recently completed a large joint project with Caterpillar, Mining3 and the Queensland Government. We’ve developed positioning technologies that they can bolt onto their underground transport AVs,” Milford said.

“The problem with the existing technologies is that GPS doesn’t work underground. If you know exactly where you are, you can improve the vehicles so they drive faster and more safely. It also helps with inventory tracking in a big, complex mine.”

The mines they’re targeting are hard rock mines, which produce much of the raw materials that go into electronics production.

“It’s great to have the opportunity to remain in Australia and develop technology to try to help Australian mining companies stay commercially competitive,” Milford said.

Inspired by nature

The technology Milford and his team developed for the mining industry was the product of many years of research in an unlikely area: rat brains.

“We like not having to reinvent everything from scratch,” Milford said. “So many of the things we want robots or AVs to do, animals and humans already do so well.

“Our research looks at how rats navigate, what sort of maps they create in their brains, and how they work out where they are in the world.”

Rats are exceptional at wayfinding, and provide an excellent model for navigation and geolocation software. Image credit: Getty Images

The team created software models of the types of navigation cells in rat brains, with the initial work led by collaborators before Milford even commenced his PhD in 2003. Through a lot of refinement and iteration, and, as Milford puts it, “lots of good engineering chucked on the top of it,” they eventually created a competent navigation system for a robot.

The product is called RatSLAM (Rat Simultaneous Localisation And Mapping). Many biologically inspired navigational systems have since been developed by other research teams around the world: BatSLAM, DolphinSLAM, and the inimitably named HamsterSLAM.

“Animals can cope with such an incredibly diverse range of environments,” Milford said. “Robots aren’t like that. Robots are highly specialised for specific environments and conditions. They don’t have anything on nature.”

Artificial intelligence and philosophy

Thanks to TV show The Good Place, the trolley problem has once again risen in the social consciousness. In the scenario, a runaway trolley is speeding down its tracks. Up ahead, there are five people tied to the trolley track. There’s an option to pull a lever and divert the trolley but doing so will switch it onto a track where one person is standing. The philosophical thought experiment asks us to decide which is the more moral course of action: letting five people die or saving them by killing one person.

It’s easy to get caught up wondering how AI in AVs would cope with such a situation.

“It’s a good hypothetical situation,” Milford said. “But for the foreseeable future, these cars can’t tell the difference between, say, an elderly person and a young adult, so there’s nothing subjective to influence its course of action.

“The solution will need to be developed collectively. A combination of industry, government and society must come up with a framework for harm minimisation. You implement it, review it regularly, and hope you can iterate towards something acceptable. But it’s never going to be perfect.”

When human lives are at stake, it’s understandable that society might be wary about putting our safety into the hands of a computer.

“There have been a lot of high-profile stories about when AI can go bad,” Milford said.

“Just like with humans, they’re a product of their input: if you bring up a human in a bad environment, they’re going to usually inherit those bad traits. Humans are inherently biased, and if an AI is left unchecked, of course it can replicate or even intensify those biases.”

The idea that AI can be a mirror for our society is an interesting one — the things reflected back at us might not always be what we expect.

“The great thing is that AI offers us an objective tool that can be very powerful in the fight against bias and discrimination. It can do things that no number of well-meaning humans can do.”

While Milford does not consider himself a philosopher, it does colour his work. “I have to be interested in the philosophy of AI in order to develop technology responsibly,” he said. “As part of that, I get to spend a lot of time with philosophers from around the world, and the conversations are very interesting, to say the least!”

“One of the things we’re responsible for as technologists is giving people realistic scenarios of how these emerging technologies could affect them, so they can plan, prepare, retool and retrain as best as they can.”

More information

Find out more about Professor Michael Milford

Explore more research at QUT’s Science and Engineering Faculty


QUT Science & Engineering

Written by

Science, technology, engineering & mathematics (STEM) news, research, insights and events from QUT Science and Engineering Faculty. #qutstem



Learning and Big Solutions (LABS) from QUT Science and Engineering Faculty

More From Medium

More on Science from The LABS

More on Science from The LABS

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade