Autonomous Helicopter Mapping

1
Autonomous Helicopter Mapping

• Andrew Ng, Mark Diel,
• Eric Berger, Adam Coates, Varun Ganapathi,
• Eric Liang, Dirk Hähnel, Rahul Biswas, Sebastian
  Thrun
• Stanford University
• Contact thrun_at_stanford.edu, ang_at_cs.stanford.edu

2
The Stanford Autonomous Helicopter

• 6-Month Goals
• Flight near obstacles, caves
• Maintenance-free hardware

Payload 14 pounds Weight 32 pounds
3
The Stanford Autonomous Helicopter
Magnetometer
GPS
IMU
802.11b
PC 104
SICK lite (3.7 pounds)
Intel Stayton

• 6-Month Goals
• Flight near obstacles, caves
• Maintenance-free hardware

Payload 14 pounds Weight 32 pounds
4
Classical Approach m-Synthesis Control
5
Our Approach Reinforcement Learning
6
Our Approach Reinforcement Learning
7
Four-legged walking
Same learning algorithm used to control complex,
very high dimensional (36D), underactuated
robots.
with Lawrence and Tal
8
Mapping (Autonomous Flight)
9
Mapping (Autonomous Flight)
10
Results (Map)
WARNING These are VRML files you will have to
edit the path to those files!
Map
Raw data
Red wall White Road Green
Vegetation Yellow Obstacle
11
Conclusions

• Autonomous flight in 11 days.
• Integrated flight and mapping.
• Next steps
• Robust flight in confined spaces.
• Coordinated ground/air mapping and navigation.