Competition
The Robot Air Hockey Challenge: Robust, Reliable, and Safe Learning Techniques for Real-world Robotics
Puze Liu · Jonas Günster · Niklas Funk · Dong Chen · Haitham Bou Ammar · Davide Tateo · Ziyuan Liu · Jan Peters
Room 353
While machine learning methods demonstrated impressive success in many application domains, their impact on real robotic platforms is still far from their potential.To unleash the capabilities of machine learning in the field of robotics, researchers need to cope with specific challenges and issues of the real world. While many robotics benchmarks are available for machine learning, most simplify the complexity of classical robotics tasks, for example neglecting highly nonlinear dynamics of the actuators, such as stiction. We organize the robot air hockey challenge, which allows machine learning researchers to face the sim-to-real-gap in a complex and dynamic environment while competing with each other. In particular, the challenge focuses on robust, reliable, and safe learning techniques suitable for real-world robotics. Through this challenge, we wish to investigate how machine learning techniques can outperform standard robotics approaches in challenging robotic scenarios while dealing with safety, limited data usage, and real-time requirements.
Schedule
Fri 7:00 a.m. - 7:15 a.m.
|
Introduction
(
Opening Remarks
)
>
SlidesLive Video |
Davide Tateo 🔗 |
Fri 7:15 a.m. - 7:45 a.m.
|
Robot Air Hockey and Other Physical Challenges: An Historical Perspective
(
Invited Talk
)
>
SlidesLive Video There have been many air hockey robots (Search for "air hockey robot" on youtube.com). I will survey ideas on how to design air hockey players, and relate them to current work on controlling robots in a variety of dynamic tasks. Two decades ago we explored manually defining primitives or skills (forehand, backhand, ...) and learning a primitive selector, first from observation, and then refining it with practice. Our view was that it is useful in learning to segment behavior into a sequence of "subroutine calls", each call having "arguments" or parameters. We chose tasks that humans do such as air hockey so we could explore learning from observation (aka learning from demonstration, imitation learning) as well as optimization-based learning approaches to learning from practice such as reinforcement learning. A key observation was that to learn from observation, the learner had to perceive in terms of primitives: segmenting behavior into individual primitives and estimating what parameters were used for each time a primitive is used. Our motivation for decomposing learning into two parts (learning skills and learning which skill to use when) is that we believed that learning a behavior selector could be very data efficient. Our approach to training the selector from observation used supervised learning, and learning from practice used model free reinforcement learning in a form that was sufficiently data efficient that all learning could be done on a physical robot rather than in simulation. One innovation we see today that was not practical 20 years ago is large scale training in simulation, transferring the learned controller to a real robot, and further learning in reality. Some current approaches to dynamic robot control pursue a conceptually similar approach of explicitly separating learning "skills" and learning to select "skills" by implicitly defining primitives by using a manually designed curriculum, and then learning a selector (in one case by distilling separate skill networks into a single network). Other approaches train on a large number of manually or automatically generated situations, and do not explicitly define a set of primitives or individual skills. |
Christopher G. Atkeson 🔗 |
Fri 7:45 a.m. - 8:00 a.m.
|
Presentation from Challenge Finalists: Air-HocKIT
(
Presentation
)
>
SlidesLive Video |
Gerhard Neumann 🔗 |
Fri 8:00 a.m. - 8:15 a.m.
|
Presentation from Challenge Finalists: SpaceR
(
Presentation
)
>
SlidesLive Video |
Andrej Orsula 🔗 |
Fri 8:15 a.m. - 8:30 a.m.
|
Highlights from the Robot Air Hockey Challenge
(
Coffee Break
)
>
SlidesLive Video |
Puze Liu 🔗 |
Fri 8:30 a.m. - 8:45 a.m.
|
Presentation from the Challenge Finalists: AiRLIHockey
(
Presnetation
)
>
SlidesLive Video |
Ante Marić 🔗 |
Fri 8:45 a.m. - 9:25 a.m.
|
Making Real-World Reinforcement Learning Practical
(
Invited Talk
)
>
SlidesLive Video Reinforcement learning offers an appealing formalism for autonomously acquiring robotic skills. Part of its appeal is its generality. However, practical robotic learning is not a perfect fit for the standard RL problem statement: from the obvious challenges with sample complexity and exploration to the deeper issues with lack of clearly specified reward functions and the practicality of episodic learning in a world that cannot be reset arbitrarily at will, making RL practical in robotics requires taking care to not only design algorithms that are efficient, but also accounting for the various practical aspects of the RL setup. This problem of "scaffolding" reinforcement learning itself involves numerous algorithmic challenges. In this talk, I will discuss some ways we can approach these challenges, from practical, safe, and reliable reinforcement learning that is efficient enough to run on real-world platforms, to automating reward function evaluation and resets. |
Sergey Levine 🔗 |
Fri 9:25 a.m. - 9:55 a.m.
|
Panel Discussion
(
Panel
)
>
SlidesLive Video |
Christopher G. Atkeson · Sergey Levine · Gerhard Neumann · Jan Peters 🔗 |
Fri 9:55 a.m. - 10:00 a.m.
|
Sponsor Talk & Award Ceremony
(
Closing Remarks
)
>
SlidesLive Video |
Ziyuan Liu · Dong Chen · Davide Tateo · Puze Liu 🔗 |