Roving eyes

Surrey Space Centre scientists are working on an ambitious two-year project to make space rovers fully autonomous and more efficient by developing a 'brain' that uses vision-based navigation techniques.

With a focus on cameras and laser-based instruments, the team aims to develop software that makes use of low-cost components and advances technology first conceived for terrestrial applications for use in planetary rovers or landers.

'The idea is to look at different vision sensors and how to make use of information [from the sensors] for navigation and guidance totally autonomously onboard the vehicle,' said Surrey's Dr Yang Gao, who specialises in space autonomy at Surrey.

'That involves sensing technology, planning and navigation techniques, and a lot of artificial intelligence techniques involved with planning the trajectory of autonomous vehicles,' she added.

'One of the work packages will focus on how we could make use of the image from the cameras and, in combination with other sensors, make the data fusion work for space applications. We also intend to make use of that information and eventually develop a dedicated navigation algorithm.

'It's like a brain, if you like, for the rover,' said Gao.

The researchers want the software to enable a rover to understand and interpret information received through its sensors and make a decision on how to move based on that interpretation.

Gao highlighted the complementary roles that different vision sensors would play in the new navigation technique.

'The camera could be used like the human eye, to see things that are further away. and you can rely on still cameras, for example, to do long-range planning.

'The laser could be used as complementary sensors for the vision — in this case it could be used for obstacle avoidance to very quickly detect and respond to obstacles within smaller ranges compared to cameras,' she said.

After acquiring this information, Gao said the rover would then be able to use onboard intelligent planning systems to decide how to get to a target location in the fastest time, with the least amount of travelling, or by using minimum resources.

'In previous missions for planetary rovers, most of them had automatic control capability, but lacked a high level of autonomy. This means that you give the rover an objective, such as 'go to that crater and have a look', but the rover does not know how to do that. It knows the final goal, but has no capability to make the decision based on the environment.

'In future space missions we want the rover to have more intelligence so that we only need to take pictures from the orbit. scientists on Earth can look at those pictures and say "that is a crater which is interesting and I want the rover to go to it". We can then rely on the rover to understand its location, understand the goal, and make the decisions along the way,' she said.

This means that if there are obstacles in the rover's planned route to a target, its autonomy will enable it to update its sensing data and revise its route. If existing rovers come across an obstacle, controllers simply tell it to stop, turn around and move in the opposite direction away from the obstacle.

'Given the advances in AI techniques in terrestrial applications, we can use a different approach. For example, the rover senses something and uses reactive control to instantaneously work around it, so it is not turning around or back, but keeps moving around the obstacle. The laser could definitely help do that, and the low-cost cameras that we put on the lower position of the robot's body could also help,' said Gao.

Currently, rovers take pictures and send the images back to Earth, where it takes a day or two for scientists to send back a command telling the rover its next move.

'The efficiency will be increased,' said Gao. 'With this autonomy, a rover could probably visit a number of interesting sites rather now, where it can barely move 100m a day.'

The Surrey team's main challenge is the reliability of robotic systems in the hostile space environment. Electronics need to go through very long testing processes before being launched into space, which means components such as expensive, high-speed processors and large memories that have not undergone vast amounts of testing are not a viable option.