Editorial: Special Issue on Calibration for Field Robotics

Every robotic system has some set of calibration parameters – scale factors, sensor locations, link lengths, etc. – that are needed for state estimation, planning, and control. Some parameters are difficult to measure by hand, for example due to the presence of sensor enclosures. Other parameters, despite best efforts during construction, will change over the lifetime of a robot due to normal wear and tear. It is often thought that autonomous systems will require many complementary sensors operating in concert, but this also means more calibration parameters to determine and manage. In the best case, incorrect parameter values degrade performance. In the worst case, they cause critical safety issues. Historically, calibration has focused on intrinsic and extrinsic parameters of single sensors in controlled environments. Intrinsic parameters are internal to the sensor, such as camera focal lengths or lidar range biases. Extrinsic parameters are external to the device, such as the mounting position and orientation of a sensor on a robot. Classic calibration methods rely on special datasets or controlled environments with expensive external measurement devices or accurately surveyed fiducial patterns for sensors to observe. Currently, we are seeing a push by researchers to tackle questions of long-term autonomy with the goal of widely deploying robotic systems into the service of non-experts for driver assistance, as personal assistants, to explore our vast oceans, and in a host of other roles. To achieve these goals, calibration parameters will have to be determined efficiently and maintained over the lifetime of a robot. Moreover, for the sake of safety and security, robots will require a certain degree of introspection in order to know when calibration parameters are no longer viable or correct. As part of this push, research is moving away from calibration that requires highly specialized environments or equipment, and towards self-calibration methods that only use onboard sensor data, or sensor data coupled with an environment model that is itself built from onboard data. This special issue of the Journal of Field Robotics includes eight papers that represent the current trends in calibration for field robots. Crucially, each of these papers addresses a problem that arises when building a robotic system with multiple sensors, or a combination of sensors and actuators. Not one of the calibration methods described relies on a high-accuracy external measurement device, and only two of the eight utilize previously-mapped environments. Two papers address the problem of online recovery of the alignment of sensors on underwater robots. Two papers address the extrinsic calibration of multiple imaging or ranging sensors by observing data captured in a static environment. One paper describes the intrinsic kinematic calibration of an actuated lidar using similar principles. The last three papers use data collected while the vehicle is in motion to calibrate either extrinsic or intrinsic-plus-extrinsic parameters.