The most anticipated robot event of the year, the DARPA Robotics Challenge (DRC), is heating up. In fact, the DRC isn't just a single event. It actually consists of three increasingly difficult competitions, and the first one is the Virtual Robotics Challenge (VRC), in which teams compete through a computer simulation of a robot and the challenge tasks. For the last month, teams from around the world have been submitting entries to the VRC qualification event, in order to earn a place in the VRC itself, and the latest results are in.
The qualifiers ended last week, and 26 teams made the cut in Track B (funded) as well as Track C (unfunded). We've got an overview of the simulation environment and the tasks that teams will need to complete in the VRC (which takes place later this month), straight from the Open Source Robotics Foundation.
We're assuming most teams want to keep their strategies under wraps, but you can watch a video from ROBIL, an Israeli robotics consortium that's qualified for the VRC, showing how they tackled some of the qualification tasks here.
To come up with these tasks, DARPA talked to disaster responders like firefighters, police, and nuclear engineers, asking them "what would you want a robot to be able to do?" In addition to things like walking over rubble and driving, a high priority was basic tool use and manipulation of hoses and valves, to (say) get cooling water into a nuclear power plant after a disaster.
As with the goal of any simulation, OSRF has put a lot of time into making Gazebo as close as possible to the real world. Maybe not so much in looks, but in how everything operates. For example, you'll notice in the video that the simulated robot doesn't move perfectly smoothly, and (if you remember), the real robot (Boston Dynamics' ATLAS) doesn't either. Ideally, teams will be able to take all the code that they've come up with as part of the VRC and have it run in exactly the same way on the actual ATLAS hardware. It won't happen this way in practice, of course, because no simulation is perfect, but OSRF has worked (and continues to work) tremendously hard on Gazebo.
Even though DARPA is footing the bill for this stuff, everyone who uses Gazebo will benefit from the improvements made to a specific set of important simulation problems, like walking. The overall goal of Gazebo is that teams shouldn't have to worry about the simulator, or even know how it works. Instead, they can focus on just writing clever code to make the robot as efficient and effective as possible, and the better Gazebo works for the VRC, the better it works in every other case as well.
Unfortunately, we're not going to be allowed to watch the VRC as it happens. The reason is that, when a given team is competing live, it gets to see only what the robot sees (first-person view), which would be the case in a real disaster scenario. In other words, teams won't be allowed to switch to a third-person view showing the robot from different positions and angles, as is commonly seen in games and simulations. If DARPA were to broadcast these different points of view, as spectators would expect to see, the team competing live could also get to see these images.
However, OSRF says that all of the VRC data will be recorded, and they'll make up some videos for us to watch after it's all over. We're hoping there will be professional gaming-style commentary from OSRF and DARPA experts (and maybe the teams themselves) to go along with it.
The VRC is scheduled to take place in a few weeks, and we'll be bringing you updates as we get them. Remember, some of these teams have a shot at getting themselves some actual ATLASes at the DARPA Robotics Challenge Trials this December. We're not exactly sure when (or where) the DRC Trials will take place, but we're for sure going to find out. And then sneak in.
[ OSRF ]