After spending an amazing week in the city of Linköping, Sweden, I am now back in the office and writing up my thoughts about the whole week.
First up was the rescue camp, the goal of the camp was to send a robot into a maze of difficult terrain and autonomously traverse it. The camp had access to 4 robots, 2 from the NIFTi project and 2 custom robots on top of a MESA robotics base. The participants split into 4 separate groups.
Here i am stopping our robot from falling over in the step field maze
My groups robot “Wowbagger” was provided by the TEDUSAR group. We broke the goal down into multiple steps:
- We would use the vicon system and the mounted kinect sensor to provide an octomap of the robots surroundings
- We would convert the octomap into an elevation map, this simplified the path planning, given that there were no tunnels.
- an obstacle map would be generated from the elevation map, denoting whether our robot could safely stand on the terrain without falling or colliding with the ground. This map was created in 3D space, 2 dimensions for the XY top down position, and 1 dimension for the orientation of the robot
- An RRT path planner to explore the maze and to only take us down paths that the obstacle map would allow.
- A small module for ensuring the servos that mount the kinect sensor were always facing a desired angle to the ground, despite the robot pitching or rolling over the terrain. This would help our mapping stage as it ensures the kinects point cloud is always of relevant areas.
- A control module that would take in waypoints generated by the path planning and tell the motors how to follow them
After 3 days, and a number of talks from Francois Pomerleau, Andreas Nuchter, Simon Lacroix and Stefan Kohlbrecher, we had learned about performing realtime SLAM, methods for quickly storing pointcloud data and methods for planning viable paths. It came time for our demonstration and unfortunately we hadn’t managed to tie the path planner into the waypoint control yet. No worries, we manually measured waypoints along the maze. So thanks to a little bit of cheating our robot was on its way. We almost made it to the end too, but after a sharp bump the onboard computer restarted and our robot lifelessly span in circles.
There were a lot of really great and relevant talks, Rescue Robotics seems to be a big and exciting field. I can only write about a few, but these are the ones I thought were the most relevant to my research.
Simon Lacroix started the conference off with his keynote about UAV and UGV coordination.
Marco Langerwisch showed a GUI for controlling multiple robots, I thought this was similar to work ORCHID has done on giving first responder’s a tablet app for controlling and allocating tasks to UAVs.
Matthew Piccoli showed his cheap, portable and modular robots that can be transformed into a car, boat or quadcopter. I thought these were a great idea, they’re cheap and fast to assemble, perfect for disaster scenarios
Paul Scerri delivered a great talk about how we should get out into the real world, get some real data and dont be afraid to lose a few robots.
Tanner Perkins spoke about a method, used at a trial in Disaster City, for coordinating UAV and UGV pilots using a mediator.
I had an amazing time and met so many great people, Hopefully I will see them again at future conferences.