(CN) – As companies working on self-driving cars continue to battle for market position, a series of limitations have conspired to keep their cutting-edge products in neutral.
However, a new framework developed by a team of researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) could allow autonomous vehicles to traverse some particularly challenging terrain: U.S. roads that are unpaved, unreliably marked or unlit.
Companies like Google currently only test their cars in major cities where they spent innumerable hours carefully labeling the exact 3-D positions of stop signs, curbs, lanes, crosswalks and offramps.
Unpaved roads, on the other hand, are often much more challenging to map – and are less frequented – so companies are not likely to develop 3-D maps for them in the immediate future. That means large portions of the United States, from Vermont’s White Mountains to California’s Mojave Desert, are not ready for self-driving cars.
With these limitations in mind, the MIT team set out to develop systems advanced enough to navigate without these maps, allowing self-driving cars to drive on roads they have never traveled on before without 3-D maps.
Known as MapLite, the framework combines simple the GPS data that one would find on Google Maps, with a series of sensors that analyze road conditions. The pairing allowed the researchers to drive autonomously on multiple unpaved country roads in Devens, Massachusetts, and reliably discern the road more than 100 feet in advance.
“The reason this kind of 'map-less' approach hasn't really been done before is because it is generally much harder to reach the same accuracy and reliability as with detailed maps,” said CSAIL graduate student and study lead author Tedd Ort.
“A system like this that can navigate just with on-board sensors shows the potential of self-driving cars being able to actually handle roads beyond the small number that tech companies have mapped.”
Ort’s paper will be presented May 22 at the International Conference on Robotics and Automation in Brisbane, Australia.
Despite considerable advancements in self-driving technology, the navigational skills of humans are still far superior to what the autonomous vehicles bring to the table.
For example, if you want to get to a specific location that you have not been to before, you likely plug its address into your mobile navigation system and consult its instructions periodically, such as when approach highway offramps or intersections.
If you were moving through the world like most self-driving cars, you would essentially be staring at your phone during your entire commute. Current systems rely heavily on maps, only incorporating sensors and vision algorithms to avoid dynamic objects like other vehicles and pedestrians.
In contrast, MapLite uses sensors for every aspect of navigation, relying on GPS data only to acquire a general estimate of the car’s location. The system first establishes both a final destination and what experts call a “local navigation goal,” which must be within view of the vehicle.
MapLite’s perception sensors then generate a path to reach that point, using light detection and ranging (LiDAR) – devices that measures distances by emitting pulses of light – to estimate the location of the road’s edges. The system can accomplish this without road markings by formulating basic assumptions about how the road will be relatively flatter than the surrounding environment.
“Our minimalist approach to mapping enables autonomous driving on country roads using local appearance and semantic features such as the presence of a parking spot or a side road,” said MIT professor Daniela Rus, a co-author of Ort’s paper.
The researchers developed a system of models that are “parameterized,” meaning they describe multiple situations that are fairly similar. For example, one model may be broad enough to decide what to do at intersections, or a specific type of road.
The system differs from other mapless driving approaches that rely more on machine learning.
Unlike mapless driving approaches that rely more on machine learning, the system is trained by data from one set of roads and tested on others.
“At the end of the day we want to be able to ask the car questions like 'how many roads are merging at this intersection?’” said Ort. “By using modeling techniques, if the system doesn't work or is involved in an accident, we can better understand why."
As part of a collaboration with the Toyota Research Institute, the team tested their system using a Toyota Prius.
MapLite still has limitations. So far, it is not reliable enough for mountain roads as it does not account for dramatic shifts in elevation. The team hopes to expand the range of roads that the vehicle can handle. Ultimately, they hope to have their system reach mapped systems’ levels of performance and reliability – albeit with a much wider range.
“I imagine that the self-driving cars of the future will always make some use of 3-D maps in urban areas,” said Ort. “But when called upon to take a trip off the beaten path, these vehicles will need to be as good as humans at driving on unfamiliar roads they have never seen before.
“We hope our work is a step in that direction.”
The project was funded in part by the National Science Foundation and the Toyota Research Initiative.
Follow @SeanDuffyCNS
Subscribe to Closing Arguments
Sign up for new weekly newsletter Closing Arguments to get the latest about ongoing trials, major litigation and hot cases and rulings in courthouses around the U.S. and the world.
