An idea that fascinates me is a robot autonomously exploring a forest and “documenting” the wildlife there. A tracked platform like this seems like it would be almost optimal, with the pan-tilt camera serving as a “hand-held” used to document wildlife. Alternatively, a small drone.
I’m curious if anything like that has been done before, or what the optimal way to accomplish it would be. My dream would be a reinforcement learning agent that learns how to optimally document wildlife, but that would require either a beefy solar powered control station for training, heavy cellular network use, or sufficient proximity to my house to transfer data to a server.
1 Like
It has been done for TV.
Generally speaking people leave cameras out in the wild and let the animals come to them. Open hardware environmental sensors are also extremely helpful. As a rule animals don’t like things that move around and make noise. I’ve come across quite a few game cameras while hiking and a number of new and endangered species have been discovered in this way. It is unclear to me what value mobility brings to the table for the added complexity. Using waterways where animals frequently congregate might be a better alternative (e.g. MBARI, a former Open Robotics customer, but for inland waterways). Or you could just spy on plants and animals from space.
Personally, I would just go for walks and contribute to the growing body of citizen science data. With any luck I’ll hit 1000 identified species this year. 
It turns out that in aggregate simple data like range, pollinators, and seasonal phenomena are extremely helpful for biologists.
4 Likes
The idea is that a robot driven by machine learning would figure out things like, “if I stay still at this location at this time, I get additional wildlife footage”. But a large part of the benefit of mobility would be for plants: finding every flower and mushroom the day it pokes out of the ground and getting daily photos of all of them. It could also get footage of squirrels and birds up in trees, ants on the ground working on ant hills, etc.
1 Like
I would say that quadruped robots might be better than the platform you sent, mainly for traversability and being less disruptive with the environment. I find the work from the Environmental Robotics Lab really inspiring.
I’ve thought about something similar and did a bit of a dive on robotics in conservation. A while back the Woods Hole Institute deployed a robot in Antarctica to track penguins. But that’s in a generally large flat area. There are some other examples I’ve seen of dressed up RC cars with cameras used for bird observation. (Sorry can’t site sources on that one as it was years ago I read about them)
And to follow up with what Kat was saying, the sound of quadcopters is so upsetting to bears that they’re deployed to chase them away from humans. So make sure you’re thinking about those kinds of factors as well.
I’m currently working on a personal evaluation of mesh_navigation to see how it can be used for navigation in forests. It’s essentially what it’s built for, and should be a good place to start.
1 Like
For is hardware, I think a four wheel diffdrive robot could also work (dependingon the terrain). The robot doesn’t have to always be in motion, it can remain at a position for some time to capture data. Additionally, some camouflage on its exterior to make it blend with its environment, and some sensors to help it detect the presence of animals would make it more effective.
Adding a robot arm with a camera mounted on it is another cool idea; it adds the flexibility to take photos from more angles, and maybe data samples can be taken with the arm.