TU-TESL

• 22

According to foreign media reports, cruise, a subsidiary of general motors, announced in June that it would delay the launch of taxi services. Previously, GM said it would launch driverless taxi service in 2019. However, the announcement of the delay did not stop the company's more than 1000 employees from iterating on the product in order to launch the service to the public as soon as possible. In fact, Hussein Mehanna, head of cruise artificial intelligence (AI) and machine learning, said the company was accelerating its development of underlying systems and infrastructure in the past few months. To this end, cruise detailed the work of the mapping team, which is responsible for building high-precision maps so that cars can automatically locate on the road. The map uses two assets - a 3D block map made of short-range and long-range lidar data and data coding labels that can provide information such as lane boundaries, traffic lights and lamp locations and curb edges. This kind of information can reduce the processing load of the car and make it focus on driving. In addition, cruise uses maps to encode the expected behavior of other vehicles based on "thousands" of interactions with a given area. Cruise's engineers used in-house production systems to test these assumptions and iterate over different versions of functionality to assess the impact of such functionality on vehicle performance, before evaluating new functionality across the map. If not used for "complex" products and operational solutions, such maps will soon become obsolete, detecting real-world changes such as construction and development projects, and sending updated information to every vehicle in the cruise fleet within minutes. Cruise launched another project called "continuous learning machine" The machine's plan aims to update the control strategy of the GM's subsidiary car (AI driver, or autopilot) more than two times a week. The goal is to enable the car to automatically detect extreme situations encountered during the driving process, and send such cases to the AI driver, so that the algorithm can improve itself. Cruise's car will encounter various obstacles such as reckless pedestrians on the road. The system of analyzing visual sensor data will focus on the pedestrian's head to identify when the pedestrian will make eye contact with the driver's side. This kind of vision algorithm also considers the attitude, through which we can try to predict the action that the pedestrians in the field of vision may take before or after crossing the road ahead. For example, when someone is detected looking at a cell phone, it can be inferred that pedestrians are unlikely to cross the road immediately, or they may be distracted. Mehanna said such algorithms are so insightful that they can even recognize people dressed as giant palm trees, which is one of the many unusual scenes cruise cars encounter when driving on public roads for many years. A year ago, the autopilot car crashed into a pedestrian case, highlighting the importance of detecting pedestrian posture. Last week, the National Transportation Safety Board issued a document saying Uber's self driving SUVs were not equipped with pedestrian recognition capabilities. In a broad sense, the real-time situation is very important in the real autonomous driving situation. Earlier this year, Cruise revealed that it was testing computer vision and sound detection AI technology to help autopilot cars deal with past emergency vehicles, which is one of the Cruise enhanced auto driving situational awareness projects. At present, Cruise continues to test autopilot in Scottsdale, Arizona and Detroit metropolitan areas, and is mainly focused on the deployment of autonomous driving vehicles in San Francisco. The size of Cruise's autopilot fleet is expanding. By June 2017, the fleet of 30 driverless cars has increased to about 130. Cruise did not disclose exactly how many autopilot cars, but the company registered 180 autopilot cars in California DMV. Three years ago, according to a document obtained by IEEE spectrum, cruise plans to deploy 300 self driving test vehicles throughout the United States.