Autonomous cars could drive into garages sooner than expected after an Australian and other researchers developed a camera that could cut thousands of dollars from their price.
When ready for sale, the artificially intelligent device is expected to cost as little as $300 compared to current technology that commands about $75,000.
Australian Catholic University associate professor Walayat Hussain announced the development on Friday, saying the smart camera could also reduce road accidents by recognising objects, pedestrians and cyclists faster than its competitors or human drivers.
The announcement comes after Tesla revealed plans to launch self-driving Cybercabs in 2026, and after US and Chinese companies including Waymo, Cruise and Baidu tested autonomous vehicles overseas.
Five researchers at the Australian Catholic University, Shanghai University and Xidian University collaborated on the AI-powered automotive technology research, which has been published by the Association for Computing Machinery.
The single camera, dubbed MonoFG, uses artificial intelligence software to distinguish between the foreground and background of a scene, Assoc Prof Hussain said, allowing vehicles to avoid objects and navigate in complex environments.
Most self-driving cars currently use several LiDAR (light detecting and ranging) systems to create a three-dimensional map of environments, even though each sensor can cost $10,000.
A Honda Legend Hybrid EX level three self-driving car, released in 2021, costs about $102,000.
"This (technology) can make self-driving technology much more affordable for the mass market," Assoc Prof Hussain told AAP.
"It proves the self-driving cars don't need expensive LiDAR arrays but a single, affordable camera which is enough when it has been trained with the right AI algorithm."
The AI-powered camera, he said, uses specialised algorithms called knowledge distillation and triple attention mechanism to identify potential hazards and, when tested against real-world data, had outperformed visual and LiDAR systems.
The MonoFG camera proved the fastest system to identify cyclists, Assoc Prof Hussain said, and could capture 18 frames per second and react to objects in as little as 0.15 of a second.
"A self-driving car using MonoFG can react almost instantly to any obstacle, traffic signal or pedestrian," he said.
"We have implemented this using the same benchmarks and datasets that are implemented in all current vehicle systems."
Future research would investigate how to scale-up the camera, he said, and seek to work with vehicle manufacturers.
Self-driving cars have been used on roads in the United States and China with mixed success.
Google-owned Waymo operates a self-driving taxi service in parts of San Francisco, Phoenix and Los Angeles, for example, but General Motors announced the closure of its rival firm Cruise in December.
Tesla has announced plans to join the self-driving robotaxi race in California and Texas as early as 2026, after unveiling its Cybercab in October.
