The clearest public head-to-head test of advanced robots is the DARPA Robotics Challenge Finals, held June 5 and 6, 2015, in Pomona, California. Robots had to finish eight disaster-response tasks within one hour, without a safety tether and with disrupted radio links to their operators. Team KAIST of South Korea won USD 2 million by completing all eight tasks in 44 minutes 28 seconds.
Key takeaways
- The Finals scored robots on the same course: eight tasks, one hour, no tether, and degraded communications.
- Three teams completed all eight tasks: KAIST (44:28), IHMC Robotics (50:26), and Tartan Rescue (55:15).
- Falls were common, and CMU's CHIMP was the only robot reported to get back up after one.
- Operators often steered foot placement by hand, which showed that autonomy lagged behind teleoperation in 2015.
- A fair "most advanced" claim names the task, the time limit, the failures, and how much a person helped.
Why one test beats a ranking list
Lists of the most advanced robot usually mix product launches, demonstration videos, and company statements. A video shows the best take of a rehearsed run. It does not show the failed attempts, the number of people nearby, or whether the robot moved on its own. A scored competition fixes those variables: the same course, the same clock, and published results for every entrant.
DARPA, the research agency of the U.S. Department of Defense, launched the program in 2012. IEEE Spectrum reported at the time that the agency named the 2011 Fukushima Dai-ichi nuclear accident as an example of a disaster that more capable robots could have helped with, and that the goal was robots able to work in spaces built for people, using tools and vehicles made for people. The challenge defined tasks, not robot shapes, so teams were free to choose legs, wheels, or both.
The course and the rules
The Finals course had eight tasks. According to Scientific American's account, they included driving a utility vehicle, getting out of it, opening a door, cutting a hole in a wall, crossing rubble or clearing debris, and climbing a short flight of stairs. The last task was a surprise that changed each day; on the second day it involved unplugging a wire and plugging it into another outlet. Each team had two attempts, and each attempt was limited to one hour.
Two conditions made the Finals harder than the earlier Trials. Robots ran untethered, so a fall could not be caught by a cable. Communication between robots and operators was also disrupted during the run. DARPA's Gill Pratt called the Finals course ten times harder than the Trials, citing the tighter time limit, the vehicle exit task, and the wireless limits, according to IEEE Spectrum's review of the Finals.
Who finished, and how fast
Teams were ranked by points, one for each task completed, and then by time. The top five, as reported by Scientific American:
| Place | Team | Points | Time |
|---|---|---|---|
| 1 | KAIST (South Korea) | 8 | 44:28 |
| 2 | IHMC Robotics | 8 | 50:26 |
| 3 | Tartan Rescue (CMU, CHIMP) | 8 | 55:15 |
| 4 | NimbRo Rescue | 7 | 34:00 |
| 5 | RoboSimian | 7 | 47:59 |
Prizes were USD 2 million, USD 1 million, and USD 500,000 for the first three places. Note the fourth row: NimbRo finished faster than any of the three winners but completed only seven tasks, so it ranked lower. A ranking by speed alone would have put a different team first.
What went wrong, and what it taught
IEEE Spectrum's review recorded many falls. IHMC's Atlas fell twice on the first day, once on rubble from an overly aggressive foot placement and once on stairs because of a software bug. MIT's Atlas fell while exiting the vehicle and broke an arm. A robot from TRAC Labs fell at the end of its second-day exit and leaked hydraulic fluid. CHIMP was the only robot that got back up after a fall.
The same review drew several lessons: operators frequently placed the robot's feet manually instead of using higher-level autonomous commands; few teams tried self-recovery; hybrid designs with wheels or tracks plus legs did best; and small errors could cause large failures. These are the reporter's conclusions from a single event, not formal findings, but they match what the scores show.
Checklist for any "most advanced robot" claim
- Is the task named, with a time limit and a pass or fail rule?
- Was the run live and unedited, and how many attempts failed?
- Did a person steer, reset, or catch the robot?
- Was the robot tested on a course it had not rehearsed?
- Do independent scorers publish the results, or only the maker?
The Finals date from 2015, and robot hardware and software have changed since. The event remains a reference point because it published results under shared rules, not because its robots are the most capable today.
FAQ
Who won the DARPA Robotics Challenge?
Team KAIST of South Korea took first place and USD 2 million in June 2015, finishing all eight tasks in 44 minutes 28 seconds with its DRC-Hubo robot.
Is the winning robot the most advanced robot in the world?
Not necessarily. It won one defined course under one set of rules. A robot built for surgery, warehouse transport, or space work would be judged on different tasks.
Did the robots work on their own?
Partly. Operators supplied much of the high-level guidance, and Spectrum reported that they often directed foot placement by hand.
Why did so many robots fall?
Reported causes included overly aggressive foot placement and a software bug, and few teams attempted self-recovery after a fall.
Related reading
- How a robot plans a route
- What is documented about the robot Sophia
- Looking for the best AI robot
- Robot technologies of the past decade
More in AI Basics.
Last verified: October 2026
Sources
- Korean Robot Takes Home $2M Prize in DARPA Challenge, Scientific American, 2015
- DARPA Robotics Challenge: Amazing Moments, Lessons Learned, and What's Next, IEEE Spectrum, 2015
- DARPA Robotics Challenge: Here Are the Official Details, IEEE Spectrum, April 10, 2012