Automated Planning for a Fleet of Micro-Air Vehicles

... doing an eight figure and have no paintball. A/C2 is on target 4, doing a drop and have 1 paintball. ... Paparazzi. Tested in simulation. Simple decision rules ...
1MB taille 2 téléchargements 299 vues
Planning

Two Strategies

Automated Planning for a Fleet of Micro-Air Vehicles Pierre-Selim Huard [email protected] Ph.D Student at ´ Ecole Nationale de l’Aviation Civile

September 18th, 2007

Paparazzi

Planning

Two Strategies

MAV Context

Paparazzi

Planning

Two Strategies

MAV Context

Size < 500mm. Time < 30 minutes.

Several vehicles. Low CPU load.

Paparazzi

Planning

Two Strategies

Outline

1

Planning

2

Two Strategies Static Dynamic

3

Paparazzi

Paparazzi

Planning

Two Strategies

Planning Input : States

States = MAVs states + Time spent + Goals achievements.

Example A/C1 is on target 1, doing an eight figure and have no paintball. A/C2 is on target 4, doing a drop and have 1 paintball. Goals : 1 and 3 are achieved.

Paparazzi

Planning

Two Strategies

Planning Input : Actions

Actions are possible decisions in Flight Plan.

Example (MAV007 Contest) Takeoff Goto waypoint Drop Eight figure Survey etc.

Paparazzi

Planning

Two Strategies

Planning Input

Mission (Goals + Constraints) Initial and Finals states. Optimisation

Example (Optimisation) Maximize the score. Maximise the score and then minimize the duration of the mission.

Paparazzi

Planning

Two Strategies

Planning Output

Plan = Sequence of Actions (almost a flight plan)

Example Takeoff, Goto T1, Eight, Goto T2, Eight, ... Land Policy = Strategy (π : States → Actions)

Example if statei then Do actionx if statej then Do actiony etc.

Paparazzi

Planning

Two Strategies

Build a policy

For each state, at each decision step → action Take into account uncertainties. State, Mission, Uncertainties → Optimal Policy.

Paparazzi

Planning

Two Strategies

Build a policy

By dynamic programming No CPU on board Large memory

Example Mission with 3 targets and 1 aircraft : 2h of CPU (on P4 3GHz) 14 000 states.

Paparazzi

Planning

Two Strategies

Real time update

Optimal plan (first guess) Update on board Fast (few seconds)

Paparazzi

Planning

Two Strategies

Implementation with Paparazzi

Open Unmanned Aircraft System (U.A.S.) Multiple aircraft ready.

Paparazzi

Planning

Two Strategies

Paparazzi

Paparazzi

Planning

Two Strategies

Tested in simulation Simple decision rules

Paparazzi

Planning

Two Strategies

Conclusions

Solution for MAVs. Experiments soon. We expect maximum score when possible.

Paparazzi