British-built brain lets UGVs navigate when GPS goes down

Every uncrewed vehicle that IDV builds, from a small four-wheeled robot through to a converted 40-tonne Terrier, runs the same British-written software, and at the heart of that software sits an unusual capability that the UK Defence Journal was able to watch in action at the company’s proving ground at MIRA in Nuneaton, a patented navigation system that lets the vehicle find its way with no satellite signal at all.

The wider system, which the company calls MACE, is the part that turns a vehicle into a robot, and Dr Geoff Davis, the managing director of IDV UK, described it in the simplest possible terms. “We’ve got MACE, that’s the brains, the technology that allows you to operate a vehicle as a robotic and autonomous vehicle, so that’s all the heavy compute and the sensors,” he said.

The navigation element is called ATLAS, it has been developed over many years at MIRA, and Andrew Maloney, the head of technology and chief engineer at IDV Robotics, explained to the UK Defence Journal that the approach owes far more to a soldier reading a paper map than it does to a conventional satnav. “Like a soldier with an OS map, if you like. It’s kind of biologically inspired in that way,” he said, before setting out how the artificial intelligence underpinning it had matured. “Around 2014 we were already using a lot of computer vision for the autonomy, and then convolutional neural networks came out and Nvidia hardware came out, and that’s led to an explosion in capability which is still continuing,” he said.

Most of the work happens before the vehicle has even moved, with aerial or satellite imagery of the operating area run through artificial intelligence that learns the ground in advance. “We take the aerial or satellite imagery, run an AI over it and classify all that terrain, then convert that into polygon boundaries, largely to compress the data, and store it in a spatial database,” Maloney explained, and the result is a memory of the landscape, its road edges, treelines and buildings, that the vehicle carries with it rather than something it has to be told over a radio link.

Once it is under way the vehicle compares what its cameras can see against that stored map and weighs up thousands of possibilities for where it must be, and during the demonstration at MIRA the team ran the GPS-denied solution alongside a conventional satellite-based one so that the difference could be seen directly. “When we’ve got edges of tracks and roads, we can be like ten-centimetre accurate, which is good enough to drive autonomously down the road, which is what we do,” Maloney said, adding that because the map is held in real-world coordinates the system produces a grid reference “just the same as a GPS would,” and explaining that the underlying mathematics is robust enough to cope with a battlefield that refuses to match the map. “It has lots of hypotheses in a probabilistic filter, and if half the stuff is destroyed or there’s new stuff, it still makes sense, just like you would if you were looking at a map,” he said.

The system can also be brought to life in places where there is no signal to begin with, Maloney described an arrangement that places the soldier firmly in control of where the vehicle believes it is. “If you’re totally denied at the start, you can initialise it by clicking on the map, saying I think I’m here, or you can type in a grid where you think you are, and then it will converge,” he said, describing a process that allows the vehicle to settle quickly onto an accurate fix without ever reaching for a satellite.

The entire design assumes the worst about the modern battlefield, namely that it will be saturated with electronic warfare and that anything which transmits will quickly be hunted down, which is precisely why the company has pushed so hard towards a vehicle that can operate without ever giving itself away. “The idea is we’re moving towards autonomy where the vehicle doesn’t have to communicate. It’s totally passive,” Maloney said. “If you emit anything, someone’s geolocating you and trying to take you out. We’re expected to be GPS-denied, intermittent comms, so we have to work with that. Contested environments are just the normal situation.”

For all the engineering that goes into the driving, the company is at pains to point out that the platform only matters because of what it carries, a philosophy Dr Davis captured in a phrase he returned to several times during the briefing. “It’s effectively like a skateboard now. The value comes from what you put on top of it,” he said, and he gave a concrete example of how the artificial intelligence pulls those payloads together into something more than the sum of its parts. “An example would be using ISTAR to automatically detect objects and send the waypoints and what it sees to other vehicles that may have a lethal payload on,” he said.

That ambition rests on the vehicle being woven into the wider command network rather than operating as an island, and Maloney said much of the team’s effort now goes into exactly that integration. “Increasingly it’s more about the system and the capability you’re getting from this uncrewed system, so we’re doing a lot of battle space management integration, and this is part of the digital backbone, the digital targeting web,” he said, explaining that a surveillance-configured vehicle can do the finding entirely on its own. “The autonomous ISTAR can go out and geolocate targets autonomously and then pump them out, like the TAC or attack system, so we can pump out that target data and images of targets onto the TAK,” he said, adding that the same connections work in reverse, so that “if you have payloads fitted, it will discover those, and then give you all the menus and functions for that payload.”

Underpinning all of it is a relationship between the soldier and the machine that the company believes has shifted with the war in Ukraine, moving from a question of saving manpower to one of saving lives. “Before Ukraine, UK MOD was fixed on not having enough soldiers, needing to reduce workload, but now Ukraine has made people realise, actually, you just don’t want to be there getting killed,” Maloney told us, and he stressed that the human stays in charge of the decisions that matter. “That teaming between the person and the system is key, having the human either in the loop or on the loop making decisions, and things like follow-me functionality where a UGV can follow you around the battlefield,” he said.

Recognising what it is looking at is a problem the company has had to solve without the luxury of real-world data, and it has turned to simulation to fill the gap, in one project teaching the system entirely inside a synthetic world that generated many thousands of labelled images of a tank across varying weather, lighting and terrain. “The system had never seen a real tank, it had only ever seen simulated tanks, but then it was able to recognise tanks,” he said, and he was clear that the perception software is the company’s own work rather than something bought off the shelf. “It’s by us. We buy the cameras and those parts, but all the software we’ve done, and increasingly we plug and play with Five Eyes, who constantly retrain the network models,” he said, describing a system that can be sent out to patrol quietly, geolocate whatever it finds and break cover only to send a short report in a low-bandwidth burst, with the resulting target information arriving as thumbnails and grid references on soldiers’ smartphones through the TAK system used by British and American forces.