Simulation of the Path Planning algorithms

The path planning module combines two algorithms - an A*-type search algorithm and Potential Field method. The A* algorithm finds the shortest collision-free path from the current (estimated) position of the vehicle toward the target position. This path provides waypoints that serve as intermediate goals for the PFM. We then apply the Potential Field method to calculate a feasible path from the current position to the farthest waypoint which is within a line of sight from the helicopter.

Simulation of the SLAM algorithm - long corridor case

This scenario includes mapping a single corridor, while moving the RW MAV from an initial position at the left bottom corner along a closed loop path with a total course length of approximately 40 m. This scenario is a good way to measure the SLAM accuracy. In most SLAM algorithms an additional loop-closure algorithm is applied to connect the previously mapped area with a new obtained data after the vehicle has returned to the initial location. The proposed method does not rely on any loop-closure algorithm to smooth and optimize the generated map. The simulations show that the accuracy at the end of the path was approximately 1-2 cm in position and extremely small in heading direction. Hence, the accuracy appears to be within less than 0.1% of the traveled 40 m distance.

Simulation of the SLAM algorithm - circle trajectory case

This scenario was to fly in a circle along a specified trajectory in a closed environment. Despite a noise level of 1.5% no error accumulation is observed as the vehicle keeps circling around and no loop closure algorithm was required to correct the map. The maximum absolute error value is approximately 9 mm for coordinates and 0.15 deg for heading. The guessed position for the SLAM algorithm in each step was generated randomly according to normal distribution with the deviation 50 mm for position and 5 deg for heading angle and with the mean equals to the prescribed trajectory points. The map quality for this case is approximately 8.5 mm.

Simulation of the full system - living quarters

This scenario represents the living quarters, where the RW MAV is initialized inside one of the rooms and is required to arrive to a goal position at the end of the map. We assume that the walls of this living quarters have a width of 100 mm or 200 mm. At the first time step, the A* algorithm estimates the path towards the target according to the first obtained scan. It causes the vehicle to turn left to go around the nearby obstacle. After 4 more steps, the passageway becomes too narrow, the updated A*-path is recalculated. This path is not valid as well, as it was discovered later on. The position and orientation estimation errors are shown to be relatively small, which guarantees the accurate vehicle navigation to the goal. All translational errors are smaller than 25 mm, while the errors in azimuth are all below 0.25 deg.

Simulation of the full system - canyon

This scenario is non-trivial and contains a dead-end sub-region. In such a case, where large changes in the heading are required, the system algorithms cause the vehicle to hover at the point of changing the heading, and then the flight resumes. Not all the obstacles are detected immediately due to the limitations of the laser range finder. The estimated A* and Potential Field paths are unremittingly checked for their validity according to the incoming data about newly detected obstacles.