Shutterstock
A vision-based system that allows off-road vehicles to drive autonomously has been developed by a team at the Southwest Research Institute (SwRI).
The technology is being positioned as a useful tool for the military, as well as space and agriculture clients. The system pairs stereo cameras with algorithms to eliminate the need for lidar and active sensors, which are typically used on autonomous vehicles meant for roads.
“We reflected on the toughest machine vision challenges and then focused on achieving dense, robust modelling for off-road navigation,” said research engineer Abe Garza.
The engineers developed a suite of tools known as the Vision for Off-Road Autonomy (VORA), which can perceive objects and model environments in real time to navigate.
Though highly reliable, lidar sensors used by many driverless cars today produce light that can be detected by hostile forces. Radar, which emits radio waves, is also detectable. With a focus on military clients, the VORA system has been designed to mitigate these issues while not having to rely on GPS navigation that can be jammed or blocked in canyons and mountains.
“For our defence clients, we wanted to develop better passive sensing capabilities, but discovered that these new computer vision tools could benefit agriculture and space research,” said Meera Towler, an SwRI assistant program manager who led the project.
The VORA technology was developed to explore planetary surfaces. In space applications, autonomous robots are limited by power, payload capacity and intermittent connectivity, and cameras make more sense than power-hungry lidar systems.
To overcome various challenges, the team developed new software to use stereo camera data for high-precision tasks traditionally accomplished with lidar. These tasks include localisation, perception, mapping and world modelling.
They developed a tool that uses a recurrent neural network to create dense, accurate disparity maps from stereovision. A disparity map highlights motion differences between two stereo images. Then the team applied an algorithm to combine sparse data from stereo image features, landmarks and readings from an inertial measurement to produce highly accurate localisation data.
“We apply our autonomy research to military and commercial vehicles, agriculture applications and so much more,” Towler said. “We are excited to show our clients a plug-and-play stereo camera solution integrated into an industry-leading autonomy stack.”
The researchers plan to integrate VORA technology into other autonomy systems and test it on an off-road course at SwRI’s San Antonio campus.
Unlike other commercially-available autonomous systems, Tesla also makes use of cameras for its Autopilot system, with each vehicle equipped with eight external cameras linked to a central computer.
The firm recently agreed to recall over two million cars in the US after regulators identified a defect with the Autopilot software.
