A Hybrid Randomized/Nonlinear Programming Technique For Small Aerial Vehicle Trajectory Planning in 3D

The use of small aerial vehicles (SAVs) such as quadrotors for inspection and surveillance applications both indoors and out is well established, but motion planning techniques are needed which can be calculated in real-time and which result in provably safe vehicle motion in tightly constrained 3D environments. Applied to small quadrotor helicopters, our planning algorithm combines Probabilistic Road Maps (PRMs) with Nonlinear Programming (NLP) to yield dynamically feasible, collision-free trajectories in nontrivial environments. Dynamics are incorporated as equality constraints in the NLP formulation, using a multiple shooting transcription. A key challenge is collision avoidance, accomplished via NLP inequality constraints. Simulation results for complex (thousands of polygons) indoor and cave environments are presented, showing complete plans computed more quickly than they can be flown.