Lidar point cloud segmentation using box prediction

A method comprising:

• receiving first data representing a cluster of points obtained using a sensor system, the cluster of points representing at least a portion of a first object and at least a portion of a second object;
• determining a first confidence metric for a first portion of the cluster of points, wherein the first confidence metric indicates a likelihood that the first portion of the cluster of points represents a first type of object;
• determining a second confidence metric for a second portion of the cluster of points, wherein the second confidence metric indicates a likelihood that the second portion of the cluster of points represents a second type of object;
• responsive to the first confidence metric and the second confidence metric exceeding a threshold value, outputting:
• second data representing the first object, wherein the second data comprises an indication of the first portion of the cluster of points and an indication of the first type of object, and
• third data representing the second object, wherein the third data comprises an indication of the second portion of the cluster of points and an indication of the second type of object;
• and
• causing a vehicle to operate in accordance with at least one of the second data or the third data.