Yes, that’s correct – that’s our team’s robot Hubo on the left, the off-the-shelf Darwin-OP from Robotis in the center, and the Boston Dynamics Petman/Atlas on the right – all in the DARPA Robotics Challenge Simulator (DRC-Sim). In the background is a computer model of our golf cart which serves as our surrogate test vehicle. Links to tutorials are given at the end of the blog. Here’s a YouTube video of the above screenshot.
Why Care? The DRC-Sim and Virtual DRC is an example of crowdsourcing – generating solutions to complex problems by leveraging the efforts of many diverse people. The DRC-Sim and the chance to be awarded a real Boston Dynamics robot provide both tools and incentive to get many solutions. From what I heard, over 100 self-funded teams registered for Track C to compete in the Virtual DRC (V-DRC) this June. It will be exciting to see what happens! With about 6-weeks until the V-DRC qualifiers in May, we’re beginning to see some cool results from Tracks B and C teams. Check out Dr. Jerry Pratt’s DARPA-funded Track B team YouTube videos!
Do You Need a PhD to Use DRC-Sim? Not exactly – but remember, the DRC is what is called “DARPA-hard” – something that is very difficult but not impossible to achieve. Past examples of DARPA-hard include landing man on the moon and radar-defeating stealth aircraft – not impossible tasks but very difficult to accomplish. The fact that there’s a $2M prize for the winner of the DRC suggests the 8-events of the Challenge are indeed DARPA-hard. Otherwise, wouldn’t anyone and everyone enter?
Frankly, those who are computer-literate can get DRC-Sim installed and running and those with Linux/Ubuntu experience can do so even faster. The difficult part is what to do afterwards! You have to write algorithms that command the robot to walk, enter a vehicle, climb a ladder, operate hand-tools and turn valves. Understanding biped walking gaits, dexterous manipulation and motion planning often demand graduate-level robotics courses. Even then, what works in simulation does not always work in the real world; things like friction, the interactions of the feet and the floor (e.g. walking on pavement is different than on sand), and dealing with lighting (e.g. shadows, glare, etc) all make the DRC “DARPA-hard”.
So Why Darwin-OP? I see the DRC as a teaching moment for building a community. The U.S. lags far behind Asia and Europe in humanoids. As such, the DRC is an opportunity for the U.S. to re-build momentum in this area of robotics and join other countries to showcase capabilities and potential. However, working with full-sized walking humanoids requires a lot of resources. In other words, the “barrier-to-entry” is high not only for general people but for research institutions like agencies, industry and academic labs.
Let’s Have a Mini-DRC! As a robotics professor (and a Track A team leader), I want students to be able to “participate” in the DRC – especially if they don’t have access or experience with real humanoids. I’m actually forming a Mini-DRC with students where the events are scaled to Darwin-OP’s size. For example, Drexel student Josh Geating custom-made a Darwin-OP sized vehicle (see figure above). See YouTube video of Darwin-OP “driving” a custom-made “car”
By having students use the Darwin-OP model, I hope they will appreciate the DRC-Sim and the disconnect between simulation and “reality”.
Tutorials: As such, I assigned Drexel engineering student Bharadwaj Ramesh who had no Linux experience go through the Tutorials on the DRC website. He authored his own tutorials (including how to install Ubuntu) and finally ended his assignment with importing Hubo and Darwin-OP models. See YouTube videos of Hubo and Darwin-OP in DRC-Sim.
As for Team DRC-Hubo, we have our own simulator (called OpenHubo) based on OpenRAVE – a well-known motion-planning simulator. If you’re interested, you can download the Hubo CAD models from Git Hub and try developing your own algorithms in either DRC-Sim or OpenHubo.
More tutorials will come. Part 2 will show how to make Darwin-OP walk in the DRC-Sim. Stayed tuned.
Please let us know what you think! Especially about Mini-DRC, or using Darwin-OP and/or Hubo in DRC-Sim or OpenHubo!