Embedded Files
- Date: 2013
- Supervisor: Masaki Takahashi
- Institution: Takahashi Lab, Keio University
For my master thesis, I have developed, implemented and performed experimental tests on the physical rover, for a fuzzy logic based navigation and terrain mapping algorithm.
The rover used to test the code was Japan Aerospace Exploration Agency’s (JAXA) test bed rover “Cuatro”. The 2 test sessions were done at the Institute of Space and Aeronautical Science (ISAS) rover laboratory, thanks to the help of prof Genya Ishigami.
The algorithm performed good during the practical tests and I have defended my MSc thesis with the highest grade.
PROBLEM STATEMENT AND OBJECTIVE
PROBLEM STATEMENT AND OBJECTIVE
- Problem to be solved
- To deal with obstacles which are not considered during global path planning due to the dead area of external sensors and occlusions.
- Mapping the explored environment.
- Objective
- To propose an efficient re-planning method considering sub-goals of the global path (their direction) and the spatial terrain properties (the state of the ground, the height and the distance from the untraversable obstacles and also the terrain roughness), in order to deal with the above problems.
Proposed Method
Proposed Method
- Global map generation from local views and position information.
- Fuzzy Controller based navigation based on local views.
CUATRO Rover
CUATRO Rover
Verification Tests on the “Cuatro” rover:
- Weight: 35kg
- 4 wheel drive
- 4 wheel steering
- SwissRanger™ SR4000 TOF camera
- Rocker suspension + independent shock absorbers
PROPOSED METHOD OVERVIEW
PROPOSED METHOD OVERVIEW
The diagram of the proposed algorithm shows the process from obtaining the short range image, through its integration in the global map (based on the robots position and attitude), up to the derivation and execution of the control variables (speed and direction) by the system.
- INPUTS: roll, pitch, yaw angles; global x, y coordinates; short range camera image
- OUTPUTS: control variables for motion execution (desired speed and orientation)
MAPPING
MAPPING
The mapping was done by taking as input the short range camera image, applying necessary transformations to compensate for the robots pose, and then integrating it in the global map. This approach considers terrain roughness and not just a simple discrimination to traversable and untraversable obstacles.
Care has been taken to appropriately weigh the old and new information when combining with the existing map data.
The figure shows how the new transformed map patch is aligned and added to the global map.
FUZZY CONTROLLER
FUZZY CONTROLLER
The figure shows how the control output is derived based on the transformed short range input image.
- OBSTACLE membership function is formed based on the distance of the untraversable obstacles
- GOAL membership function is a simple triangular function with its peak at the direction of the goal.
The above two are combined to get the final output control variables – speed coefficient and heading direction.
TESTING ON “CUATRO” Rover
TESTING ON “CUATRO” Rover
TEST 1 – follow sub-goals and avoid obstacle
TEST 1 – follow sub-goals and avoid obstacle
TEST 2 – manage the occluded terrain
TEST 2 – manage the occluded terrain
Page updated
Google Sites
Report abuse