Tuesday, February 26, 2013

Reactive obstacle avoidance for rotorcraft UAVs_Hrabar

Reactive obstacle avoidance for rotorcraft UAV
Stefan Hrabar
2011 IEEE/RSJ International Conference on Intelligent Robots and Systems September 25-30, 2011. San Francisco, CA, USA

Abstract- We present a goal-directed 3D reactive obstacle avoidance algorithm specifically designed for Rotorcraft Unmanned Aerial Vehicles (RUAVs) that fly point-to-point type trajectories. The algorithm detects potential collision within a cylindrical Safety Volume projected ahead of the environment. An expanding elliptical search is performed to find an Escape point: a waypoint which offers a collision free route past obstacles and towards a goal waypoint. An efficient occupied voxel checking technique is employed which approximates the Safety Volume by a series of spheres, and uses an approximate nearest neighbor search in a Bkd-tree representation of the occupied voxels. Tests show the algorithm can typically find an Escape Point in under 100 ms using onboard UAV processing for a cluttered environment with 20 000 occupied voxels. Successful collision avoidance result are presented from simulation experiments and from flights with an autonomous helicopter with stero and laser range sensors.


Their ability to take-off and land vertically, hover in place or traverse slowly make this type of vehicle suitable  for applications such as surveillance, search and rescue, structure inspections or package delivery to areas without airstrips.

Two layer obstacle avoidance: Global path planning and local obstacle avoidance

Two ways of reactive obstacle avoidance for UAV:

    1. Map-based
    2. Mapless

Mapless approaches have been popular for smaller platforms
with limited processing power or where absolute range 
measurements are not available. These approaches typically 
rely on vision-based cues such as optic flow [3] [4] or vision based 
relative navigation [5]. Laser-based mapless reactive 
avoidance has also been demonstrated on our UAV platform 
by Merz and Kendoul [6]

还有一种方法是只 plan而不avoid, 通过快速重新plan来避开障碍物。

1. Rapidly Exploring Random Trees( RRTs) [8]
2. Probabilistic Roadmap(PRMs) [9, 1]
3. RRT+PRM [10]
4. Anytime Dynamics(A*) [11]

本文提出的观点是 在 local reactive 的时候考虑global 的地图信息,以找到一个中间的waypoint,在这个waypoint 之前和之后都没有障碍。

No comments:

Post a Comment