# ICP results as command in odometry motion model

im going through the odometry motion model proposed by Thurn in Probabilistic Robotics famous book and i have some questions concerning the kind of command that we can use. I will explain more my problematic below :

The odometry motion model from Thurn is :

So the motion of the robot depend on the previous state AND the three basic motions described as :

This three basic motions are computed from a relative position acquired from sensor readings, but in most books,publications that i read, this sensor readings refer to odometry(encoders) readings. So im wondering if i can use for example the relative positions given by ICP scan matching methods in the odometry motion model described above OR i have to use it directly as motion information in the prediction step of particle filter for example.

In "An Improved Technique for Robot Global Localization in Indoor Environments" article, they use ICP results to compute motion of robot but i dont understand if it used in an odometry motion model or it used directly.

NB : I tried to use ICP results as command in odometry motion model of AMCL in ROS and it works perfectly but its not really clear for me because when publications mention odometry motion model, they always refer to encoders readings.

Thank you

• you can use sensors such as LIDARs to perform your scan and match the points you get from the scan you can then match them to a known map to perform ICP and use that to localise within a known environment. There will be errors associated with this approach, you can add the encoder data into your model to reduce the error. Jul 30, 2020 at 2:57
• Hi @JJerome, Thank you for the reply, i already try that and it works perfectly but my question concern the correctness of using the relative position derivation by ICP into the odometry-motion model that i referred in my original post. Sorry for my english, maybe its not clear Jul 31, 2020 at 16:12
• Using Encoder readings alone is not the best way to solve this problem. Wheels can spin freely on slippery and lose gravel surfaces that might throw off your calculations unless your control system is able to detect slips and can correct itself. It is best to use both the Lidar scan and the encoder reading to perform error compensation. PS your English is fine! :) Aug 3, 2020 at 6:29
• Thank you for the clarification Jerome, appreciate your help, im planning to fuse the information from both sources with an EKF for example ! Aug 4, 2020 at 15:56