I was wondering if members of the ROS community were aware of any past/ongoing open-source robotics projects around humanitarian demining?
i.e projects with the goal of designing and building a software/hardware architecture for robots (Rovers,UAVs) to perform autonomous or semi-autonomous landmine detection and/or removal.
It seems like the kind of application that could make great use of the ROS framework and maybe also some open-hardware design, so I am surprised to see relatively few references in Google :
I was wondering what were the main obstacles to creating such a project :
Cost of hardware too high for individual contributors to build their platform?
Difficulty to reproduce a proper testing environment ?
General lack of interest for humanitarian robotics?
As a ROS user with experience in navigation, this is the kind of project that I would be glad to dedicate some spare time to, so I just wanted to start by collecting some general feedback. What are your thoughts?
When ROS was first released, I was developing software for Cornell Minesweeper, which was a set of robots devoted to just this problem. Surprisingly, almost 5 years after we shut the team down, the website is still up: http://minesweeper.engineering.cornell.edu/
We used an early form of the ROS navigation stack, a GPS, an old SICK lidar. We tested out at Ft. Belvior in VA. Our biggest problem, failure on test day. Everything ran great the night before the demo. Then we got out in front of the military, and the robot refused to move. Cornell cut funding after that kind of failure, and we just all went our separate ways. I definitely think it’s a good problem to go after, and I’d be more than happy to share my experience in designing and developing the different Minesweeper robots.
As far as why you don’t see any calls for remote developing, there were certainly a great number of hardware specific issues, developing drivers/firmware/interfaces etc. Hardware was expensive, ~20k, and that was for just a metal detector, some of the fancier landmine detection methods cost 100k+. Some of this could be alleviated by allowing network access to sensors, or possibly by bagging lots of data if you need signal processing type work. Testing environment is dependent on the particular sensor type, in our case, we just buried nails in dirt. Lastly, I’d say there is a desire to help with humanitarian causes, at our largest we were a team of just under 30 undergrads. Not the biggest team at Cornell, but a good medium sized team. Smart and enterprising people, some of whom went on to found Kiva Systems.
Thanks Barrett, and sorry for the late response. That looks like a very interesting project, and sharing some of that experience with the community is a big help for sure.
I am interested in collecting more feedback before actually starting any work because I will have only limited spare time to dedicate to it myself over the next few months.
Since the hardware cost and development seems to be a major issue, I am wondering if such a project could be broken down in sub-projects, abstracting parts of the hardware wherever we can. Maybe something like :
Platform development : defining a software architecture to interface and control some common rover platforms for outdoor navigation. I don’t know how heavy mine sensors are or what power they consume but that would obviously impact the choice of the platform, in addition to the usual list of specs for rovers.
If available ROS-ready platforms can do the job, we would have to consider whether they are affordable enough for a community of developers starting with no external funding. To interface with much more affordable “crawler” RC cars, we could try to collaborate here with DIY communities like ArduRover ( http://diydrones.com/group/ardurover-user-group).
Mine sensing : developing ROS drivers to interface with one or several common sensors for mine detection. This area is a big unknown for me, so I would be glad to hear more from users who have experience with such sensors, and maybe start listing some commercially available sensors with their characteristics.
Autonomy : outdoor localization, mapping and planning. For development, this part should be able to use a simple model of the mine sensor(s), and a mix of simulations and field tests with more basic platforms, like cheaper RC cars.
Some suggested items to move forward would be :
Collecting more feedback from the ROS community, investigating collaborations with other communities
Refining the breakdown of subprojects above
Listing commercial mine sensors and their characteristics
Defining typical test scenarios and reproducible test environments. Listing the typical specs of a “demining rover”
Listing related online resources
Whether or not you have experience in the field, if you are interested in such a project, please share your ideas!