Development of A Mobile Robot System

Autonomous operation of mobile robots is crucial for many applications. We details the development of a mobile robot system, concentrating on smooth trajectory generation and 3D sensing.

Smooth trajectories are trajectories which are continuous in curvature. They limit wheel slippage which can affect localisation and control, and can reduce wear on motors. Smooth trajectories also produce robot motion which is is easily predictable by humans, thus increasing safety. 3D sensing is necessary to operate in arbitrary environments.

The system produces two dimensional path plans from point to point within a grid based map. This path is discrete thus the robot has trouble following it, while still avoiding obstacles both in the map and current sensor readings. We develope a method using fourth order curvature polynomials for generating smooth trajectories to a local goal while avoiding obstacles.

Well known cubic polynomials can be used to give continuous smooth trajectories. To avoid obstacles we introduce a fourth order term, desribing cost to obstacles in the sensor view. Starting with an intitial cubic polynomial, this term drives the trajectory away from obstacles in a form of gradient descent.









Trajectory generation fails when:
- goal outside a limited local range
- configuration of obstacles either causes local minima or impossible tomodel
Future work: continuous path planning

Laser pans up/down with sinusoidal motion. Laser range finder scans 80 times in one sweep with 181 points for each horizontal scan. One full 3D scan takes about 1 second. Translational and rotational motion accounted for.



The 3D data points are segmented into planes using the random two point sampling algorithm. Planes can be found even when robot is moving (~0.3ms). Further work is to extract polygon regions from planes for 3DSLAM and object recognition applications.