Motion compensation (often referred to as “deskewing”) is a technique used to correct for distortions in the pointcloud when collecting Lidar data while the sensor is in motion. Because mechanical spinning Lidar sensors like the ouster os1-64 do not capture the entire 360-degree environment at the same instant in time, thus distortions manifest during both the translation (linear movement) and rotation (turning) of the sensor.
When a sensor is mounted to a moving vehicle, the physical position of the sensor changes between the time the scan starts and the time it finishes (typically 100 milliseconds for a 10 Hz spin rate). The collected lidar data will have a distortion effect similar to the one shown in the following image:
If raw, distorted Lidar data is fed into a SLAM (Simultaneous Localization and Mapping) algorithm without correction, it severely degrades the scan matching process. The algorithm will struggle to align consecutive frames, resulting in “ghosting,” duplicated geometry, and a degraded final map.
A motion compensated Lidar data of the previous example will ideally similar to the following image:
While translation stretches the data, rotational distortion can be even more destructive. If the vehicle makes a sharp turn while the sensor is spinning, straight lines (like walls or building facades) will appear curved, warped, or split as shown in the following image:
Methods of Compensation
To fix these distortions, we must estimate where the sensor was at the specific timestamp at which each laser pulse was fired.
1. The Constant Velocity Motion Model
A common baseline approach is the constant velocity model. This model assumes that the sensor is moving with a constant velocity between subsequent poses and interpolates the sensor’s position and heading over time.
Pros:
- Works well for predictable motions (no sudden changes in velocity or direction).
- Does not require additional hardware (IMU).
Cons: Fails completely during erratic movement such as sudden braking, hitting a pothole, or making sharp turns.
2. IMU-Aided Motion Compensation
To overcome the limitations of the constant velocity model, an Inertial Measurement Unit (IMU) can be used to detect the motion direction and speed and can be used to aid with the motion compensation process. An IMU measures linear acceleration (via accelerometers) and angular velocity (via gyroscopes) at much higher frequencies than the spin rate of a Lidar sensor (typically in the range of 100 Hz and 1000 Hz). This provides inputs to the motion compensation process that greatly improves its accuracy.
Ouster FW 3.2 and SDK 0.16
Ouster 3.2 firmware improved the current integration and outcome of the builtin IMU providing a higher frequency of IMU samples that is readily time-synced with the Lidar data.
The Ouster SDK 0.16 SLAM and Localization modules mathematically integrate the IMU data (linear acceleration and angular velocity) over the period of the LidarScan to provide a precise pose estimate (position and orientation) for every timestamp in the LidarScan.
The Ouster SDK expose these motion compensation capabilities through both the CLI and Python API.
This offers an improved scan registration process and a more accurate map.
Using Motion Compensation via CLI
The Ouster SDK CLI provides commands to apply motion compensation to recorded data.
To specify the motion compensation method during slam use the --deskew-method option:
ouster-cli source $SOURCE_URL slam --deskew-method $DESKEW_METHOD viz
The --deskew-method option has one of the following:
-
"auto": [default] Automatically select the method based on the sensor firmware version. -
"constant_velocity": Use the constant velocity motion model. -
"imu_deskew": Use the IMU-aided motion compensation (requires FW 3.2). -
"none": Disable motion compensation.