MIT develops new technologies: Driverless cars can also be navigated without maps: CSAIL, an MIT-affiliated laboratory, equipped the driverless Toyota Prius with a new system that can be used on rural roads.
To date, many unmanned cars tested on the road have either relied on highly detailed 3D maps or the system has allowed vehicles to navigate in well-characterized urban or highway environments. However, the vast majority of US roads have not been accurately Maps into three-D maps, or can't reliably provide consistent lane markings. Therefore, the MIT team at the Computer Science and Artificial Intelligence Laboratory (CSAIL) has developed a method that allows driverless cars to read And forecasting the local environment, rather than relying on 3D map data.
CSAIL graduate student Teddy Ort said: 'This 'no-map' model has never been tried before because it is often difficult to have the accuracy and reliability of detailed maps. Systems like this can be Navigating on-vehicle sensors shows the potential of driverless cars that can handle road conditions beyond the maps drawn by technology companies.
The researchers drafted a report describing the system, which will be submitted later this month at the International Conference on Robotics and Automation (ICRA) in Brisbane, Australia. To test the system, the researchers were none The Toyota Prius is equipped with a lidar, sensors, and MapLite. This Prius can 'see' road conditions 30 meters ahead, successfully driving on a number of unpaved country roads in Massachusetts.
The researchers stated in the report: 'This means that it can make the car run at speeds of over 100 kilometers per hour, and if the method can be parallelized and implemented on the GPU, the car can travel faster.'
Olt explained that their system is also different from other 'no map' models, which use machine learning to train the system. They use data from a set of roads, then test their patterns and apply them to other roads. In contrast, MapLite tries to develop a model of the conditions that the driverless car may encounter and then informs it of its behavior, which is not much different from the way human drivers drive in familiar situations.
Olt said: 'In the end, we hope to be able to ask more questions about cars, like 'How many roads are being merged at this crossroad?' By using modeling techniques, if the system is not working or an accident occurs, we can better Understand why. '
Although a system like MapLite can open up more directions for the development of driverless cars, it still has a long way to go before it is put into use. Researchers claim that their systems cannot explain the changes in altitude, so now it is impossible to cope with the mountain road. Say: 'I think the future driverless cars will always use 3D maps in urban areas. When we are driving on remote rural roads, these vehicles need to be able to respond to them as if they are on strange roads. We hope our efforts will take a big step in this direction.'