Following an accident involving one of its robotaxis, General Motors’ self-driving business and the US auto authority said on Wednesday that Cruise is recalling 950 driverless vehicles from US roads and may remove more.

According to a notice posted on the National Highway Traffic Safety Administration (NHTSA) website on Wednesday, the reason for the recall is that the Cruise Automated Driving Systems (ADS) software’s collision detection subsystem may react erroneously after a collision.

A hit-and-run motorist struck a woman in San Francisco last month, pushing her into an adjacent lane where she was struck again by a Cruise robotaxi that was unable to stop in time.

According to the NHTSA website, Cruise has updated the software on all of its supervised test fleet cars over the air. Before going back into service, all damaged driverless cars will also be fixed, according to the statement.

A fresh investigation of the autonomous vehicle’s reaction to the collision led to the voluntary recall of GM’s Cruise, and the firm said in a statement to Reuters last week that it may file for more recalls.

“Today we have issued a voluntary recall of part of our AV software based on a new analysis of our AV’s post-collision response on October 2nd,” Cruise said.

After the California Department of Motor Vehicles ordered the robotaxi operator to remove its driverless vehicles from state highways, Cruise said late last month that it will stop all operations globally.

Regarding the safety of its vehicles, Cruise is the subject of many government inquiries. In two instances, the robot cars seemed to fail to stop for people crossing the street.

With facilities in Phoenix, Arizona; Houston, Austin, and Dallas, Texas; and Miami, Florida, Cruise is competing with other companies, including Alphabet’s Waymo business, to be the first to sell robot vehicles.

In order to run software, autonomous automobiles need sensors, actuators, sophisticated algorithms, machine learning systems, and potent CPUs.

Using a range of sensors located throughout the vehicle, autonomous automobiles build and update a map of their environment. Radar sensors track the whereabouts of neighboring cars. Video cameras follow other cars, watch for pedestrians, read road signs, and detect traffic signals. 

Lidar, or light detection and ranging, sensors use light pulses to detect and identify lane markers, measure distances, and detect road borders around the automobile. When parking, the wheels’ ultrasonic sensors identify other cars and obstacles.