2019 Competition Results

This year we had 7 participants, 3 improved versions of some of the best agents of the past few years, the defending champion and 3 new agents. We had improved versions of AngryHex and IHSEV (now called Orpheus), the two teams with the most experience who participated in all competitions since 2013. IHSEV was the finalist in 2016 and 2017, AngryHex made it to the semi finals in the past. BamBirds, the winner from 2016 and finalist from 2018, participated again with a much improved version. They are still the only agent that plans ahead. As every year, the defending champion participated with an unmodified version. In 2017, Eagle's Wings was the surprise winner, developed by Tian Jian Wang, an undergrad student from the University of Waterloo. Eagle's Wings won again in 2018, so the agent we run is again the unmodified version from 2017. The new agents this year were Shiro from NIAD in Japan, SimbaDD from TU Dresden who participated already in 2016 with Seabirds, as well as MMC from FEI in Brazil. It happened quite a few times in the past that newcomers won the competition, so anything was possible. 

At every round, the agents have 30 minutes to solve 8 new levels that we design ourselves. We have not played any of the agents on the new competition levels and they are unknown to all participants. For every match, all agents can see the current high scores per level for all agents of the same match. and can use this information to select which level to solve next. Levels can be solved again in any order until the time is up. What was new at this competition were the fancy new laptops we bought from a grant from the Artificial Intelligence journal. Each agent runs on an individual laptop and can use its full computational power. We expected that particularly computation intensive agents would benefit from the faster laptops. 

The first quarter final was already very exciting. Angry-Hex and Orpheus were solving some levels and took the lead. SimbaDD started slow, but solved level after level until the were ahead of the other two competitors. They ended up solving all 8 levels, which was very impressive, and won the first quarter final easily. We expected Orpheus to perform particularly well as their simulation approach would benefit from the faster laptops, but surprisingly they were last and at the risk of elimination. Angry-Hex did much better than last year and were second. Overall we saw some very impressive shots by the different agents, but it wasn't yet clear if any of those agents will qualify for the semi finals, as only the 4 best agents overall will move on, no matter in which quarter final they played. 

The second quarter final was exciting as the two past champions were competing against each other. Last years winner Eagle's Wings and the 2016 winner BamBirds. In addition we had Shiro which did very well in our initial compatibility tests, plus MMC which was based on Deep Reinforcement Learning, an approach very successful in many other games. We were expecting some very strong performances. Angry-Hex and Orpheus had all reasons to be worried. BamBirds delivered, they showed a very impressive performance and managed to solve all 8 levels, but with a slightly lower score as SimbaDD. The big surprise, however, was Eagle's Wings. They only solved a few levels and ended up behind Orpheus. The defending Champion was out, the first big sensation of the competition! Shiro was surprisingly weak, we expected more. Like all Deep Learning agents in the past, MMC also didn't do very well and ended the competition in seventh place. All three teams from the first quarter final qualified for the semi final, while only BamBirds from the second quarter final. We couldn't believe it. Also, Orpheus (former IHSEV) did it again. They were in all semi finals since 2013. 

The semi final was without surprises. The levels were now more difficult than in the quarter finals. Just like in the quarter finals, SimbaDD and BamBirds were much better than Angry-Hex and Orpheus, halfway through the semi-final the two leaders were only 20 points apart. Thanks to one high scoring level they solved, BamBirds was ahead this time, again with some spectacular shots. Angry-Hex ended the competition in third place, while Orpheus was fourth overall. 

With their excellent performance throughout the competition, it was not clear who would win the Grand Final of the two best agents BamBirds and SimbaDD. The 8 levels we used were again harder than the semi final levels, some of them were too hard and couldn't be solved by any of the two agents. But still, both agents solved a number of levels very well. The Grand Final was exciting until the very end. BamBirds was ahead by just 30,000 points. If SimbaDD had solved just one more level, they would have won. They tried very hard in the last few minutes, but couldn't catch up with BamBirds. 

Therefore: BamBirds is the new AIBIRDS Champion!!! 

Congratulations to Diedrich Wolter from the University of Bamberg in Germany and his large team of students who won the title for the second time. Very well done! Diedrich was also the only one of all participants who attended in person, while the others participated remotely. We are sure it was well worth the trip to Macau.

That was the end of a very exciting competition. We saw many improvements and  amazing shots. We all felt like AI is getting closer to human performance and that this year would be the first time where AI has a realistic chance to beat humans in our annual Man vs Machine Challenge. The four semi-finalists qualified for the Man vs Machine Competition the next day and it promised to be very exciting. 

We hope to see many improved agents and many new agents at our next competition in 2020. Angry Birds remains a very challenging problem for AI. We are still waiting to see an exceptionally good deep learning agent. We encourage and challenge all members of the AI community to take on this problem and to develop AI that can successfully deal with a physical environment. See you in 2020! 

Jochen, Katya, Hua and Peng. 

All scores of the competition can be found in the following tables: