Sensor platform

One or more non-transitory computer-readable media storing instructions executable by one or more processors, wherein the instructions, when executed, cause the one or more processors to perform operations comprising:

• receiving sensor data collected from a sensor platform which includes two or more sensors comprising at least a first lidar sensor and a second lidar sensor, and which is removably mounted as a retrofit to a first vehicle, the sensor data collected while the first vehicle is operating in an environment at a first time;
• and
• training, based at least in part on the sensor data, a machine-learned model deployed on a second vehicle operating at a second time, the second vehicle an autonomous vehicle, wherein:
• the second vehicle includes two or more sensors comprising at least a third lidar sensor and a fourth lidar sensor, the two or more sensors corresponding in modality, location and orientation to the two or more sensors of the sensor platform, the second vehicle comprising one or more mounts for receiving the two or more sensors of the second vehicle, the one or more mounts integral with the second vehicle, wherein:
• the third lidar sensor is positioned at a height above a road surface corresponding to a height of the first lidar sensor of the sensor platform when the sensor platform is mounted on the first vehicle;
• and
• the fourth lidar sensor is positioned at a height above the road surface corresponding to a height of the second lidar sensor of the sensor platform when the sensor platform is mounted on the first vehicle;
• and
• training the machine-learned model comprises training the machine-learned model to determine outputs based at least in part on a first sensor reading from a first sensor of the sensor platform such that the machine-learned model is configured to generate the outputs based at least in part on a second sensor reading from a second sensor of the second vehicle, wherein the second sensor of the second vehicle corresponds in modality, location and orientation to the first sensor of the sensor platform.

A method, comprising:

• collecting, by a sensor platform mounted as a retrofit to a first vehicle operating in an environment at a first time, sensor data generated by a first group of sensors mounted on a sensor platform, wherein the sensor platform is removably mounted on a roof of the first vehicle, a first detection region associated with the first group of sensors extends horizontally in all directions from the first vehicle, and the first vehicle is operable in a non-autonomous manner under the control of a human occupant of the first vehicle, the first group of sensors comprising at least a first lidar sensor and a second lidar sensor;
• training, based at least in part on the sensor data, a machine-learned model deployed on a second vehicle, the second vehicle an autonomous vehicle, wherein:
• the second vehicle includes a second group of sensors comprising two or more sensors each corresponding in modality, location and orientation to one of the sensors of the first group of sensors, a second detection region associated with the second group of sensors extending horizontally in all directions from the second vehicle, each of the second group of sensors mounted within a different respective sensor pod, each respective sensor pod comprising one or more sensors, each respective sensor pod being coupled to a respective mount on the second vehicle, wherein:
• the second group of sensors comprises at least a third lidar sensor and a fourth lidar sensor;
• the third lidar sensor is positioned at a height above a road surface corresponding to a height of the first lidar sensor of the first group of sensors when the sensor platform is mounted on the first vehicle;
• and
• the fourth lidar sensor is positioned at a height above the road surface corresponding to a height of the second lidar sensor of the first group of sensors when the sensor platform is mounted on the first vehicle;
• and
• training the machine-learned model comprises training the machine-learned model to determine outputs based at least in part on a first sensor reading from a first sensor of the first group of sensors of the sensor platform of the first vehicle such that the machine-learned model is configured to generate the outputs based at least in part on a second sensor reading from a second sensor of the second group of sensors of the second vehicle, wherein the second sensor of the second vehicle corresponds in modality, location and orientation to the first sensor of the sensor platform of the first vehicle;
• and
• operating the second vehicle in the environment at a second time using the trained machine-learned model, the second time after the first time.