A Xenomai Linux based robot from Korea’s Team KAIST called the DRC-Hubo won the $2 million DARPA Robotics Challenge Finals, one of only three bots to complete the course on time.

Judging by Silicon Valley’s reigning “Failure rocks!” mantra, this week’s DARPA Robotics Challenge Finals held earlier this week in Pomona, California, was a resounding success. Of the 23 teams that entered the event, which simulated a disaster response scenario, only three accomplished all eight major tasks in the allotted eight hours. The pit crews were kept busy with a field littered with falling robots.

Team KAIST’s first place DRC-Hubo
(click images to enlarge)

The winner — South Korea’s Team KAIST — brought home $2 million in prize money with its DRC-Hubo humanoid bot, which runs real-time Xenomai Linux. (Eighth place Team DRC-Hubo from UNLV also used a DRC-Hubo design.) The $1 million second place prize went to the Institute of Human and Machine Cognition (IHMC) Robotics in Pensacola, Florida, which fielded a “Running Man” version of the Atlas humanoid robot from Google’s Boston Dynamics.


Carnegie Mellon’s Team Tartan Rescue CHIMP
(click image to enlarge)

Third place ($500,000) was held by Carnegie Mellon’s Team Tartan Rescue and its CHIMP robot. Judging from a job posting last year for an embedded Linux software engineer, the CHIMP looks like it may be a Linux-enabled robot as well.

Unlike some of the contenders, the top three winners were all bipedal humanoid robots, although the CHIMP and DRC-Hubo are really more like Transformers. The CHIMP can travel on all four with its arms, as well as walk and roll on rubberized tracks, and the DRC-Hubo can switch from walking to rolling along on motorized wheels built into its knees. According to an IEEE Spectrum story, the combination of bipedal and wheeled modes helped Team KAIST remain stable as it completed all eight tasks in only 44 minutes, 28 seconds.

DRC-Hubo spinning 180 degrees on its torso (left) and resting on its motorized roller knees
Source: Image on right from IEEE Spectrum

 
The DARPA challenge required the robots to adapt to a wide variety of tasks that most humans could accomplish fairly easily. They were required to drive a vehicle, get out, and then open a door and pass through. The robots then negotiated a debris field, used a tool to cut a hole in the wall, and reached through to open a valve. The bots then negotiated another obstacle course with debris and pipes, and inserted a plug into a receptacle. Finally, they climbed some stairs and performed a “surprise manipulation task.”

IMHC Running Man (Atlas)

The CHIMP was the early leader, with MIT’s version of the Atlas in hot pursuit. However, the CHIMP missed a mark and ran into a barrier on the final run, and MIT’s Atlas broke an arm, enabling Team KAIST to win the day, followed by the Team IMHC’s Running Man.

Most of the bipedal robots fell to the ground when getting out of the vehicle, running the drill, or opening a door. Thanks in part to its strong arms, the CHIMP was the only bot that was able to regain its feet without human help.

Team KAIST’s DRC-Hubo remained stable throughout thanks to motorized wheels for the knees, stabilized by casters on the toes. The design also improved speed on unencumbered stretches of the course.

The 180cm tall, 80-kilogram DRC-Hubo cost between $500,000 and $1 million to build, according to IEEE Spectrum. The robot, which has been under development since 2002, was principally designed by Team KAIST leader Jun Ho Oh, a professor of mechanical engineering at the Korea Advanced Institute of Science and Technology (KAIST). A KAIST spinoff called Rainbow Co. was instrumental in the design, as well as RCV Lab, which contributed computer vision technology.

The semi-autonomous design uses Xenomai real-time Linux, which has been adopted for other bipedal robots such as Oregon State University’s ATRIAS. In addition, the software stack incorporates a custom-made a motion control framework called PODO, as well as the Gazebo simulation environment.

Other keys to Team KAIST’s success include the DRC-Hubo’s powerful, air-cooled motors, which enable 31 total degrees of freedom. Its 33 motors feature customized drivers with special amplifiers instead of conventional feedback controllers. The power supply for the motors had a supercapacitor-based backup in case of failure.

Other innovations included a torso that could rotate 180 degrees, which came in handy while driving, cutting, clearing rubble, and climbing stairs, says IEEE Spectrum. The 7-DoF arms, each of which can carry 15 kilograms, are longer than usual, and fitted with flexible, “adaptive gripper” hands.

DRC-Hubo puzzles out “surprise manipulation task”

The vision system on the DRC-Hubo is simpler than those on many of the contestants, such as the Atlas. Yet, the robot performed vision-intensive tasks such as grasping and manipulating objects that slowed down more advanced vision systems.

For the most part, the robot relied on regular cameras, and only used the LIDAR imaging device when needed, says IEEE. The simpler design helped reduce the size of the head, improving stability, among other benefits. The success of this simpler camera-based design adds weight to the argument that improving image recognition algorithms will enable robots and drones to navigate largely by regular cameras rather than more expensive imaging systems.

Team KAIST’s DRC-Hubo performs final “surprise manipulation task”
Source: IEEE Spectrum, YouTube