What tasks should robotics researchers focus on?

This workshop is a spiritual successor of the benchmarking workshop at CoRL last year, where one of the main takeaways was that academic roboticists do not have a clear, commonly agreed upon set of robotic tasks to focus on. 

This workshop will benefit the robot learning community in several ways. First, we hope to replicate the progress made in the computer vision and NLP communities in robotics (which has mostly been driven by specialized tasks and metrics) by coming to a resolution on specific robotics tasks that the robot learning community can work on together. 

Second, we would like to understand why the translation of academic advancements in robotics into practical commercial applications remains a challenge, resulting in a fragmented landscape where industrial researchers often resort to creating customized contraptions and optimizing the surrounding environment to achieve simplified solutions (sometimes without any learning, even). While we have several amazing accepted papers at CoRL, the reality is that very few papers successfully transition into tangible real-world applications. Can this be addressed by finding tasks that are suitable for both academic research and commercial applications?

Lastly, we would like to discuss the recent rise of general-purpose humanoid robot companies. As opposed to industrial and/or collaborative robots, which have been focused on automating specialized tasks like insertion, tending, pick and place, general-purpose humanoids have the potential to solve a wider array of tasks. This will significantly influence the next generation of robot hardware available to robot learning researchers. How should the availability of general-purpose humanoids inform the tasks that researchers focus on moving forward?

Below are the discussion questions we would like to focus on:

• How can industrial robotics companies help academic labs and vice versa?

• What are some tasks that we can potentially standardize?

• Should we work on general-purpose humanoid vs specialized robots? 

• In academic robotics, why does everyone invent their own tasks?

• Academics, do we want to see our research used in industry, or just write papers?

• As academics, how do we balance problems between home vs industrial robotics? 

• How do we track progress in these settings? 

• The robot learning community proposes new benchmarks and common tasks all the time, yet adoption by other robot learning researchers seems low. Why is this?