2 of 4

Summary

How will these algorithms scale up? It seems some of them may, through lookup tables.
Will they eventually make it out of the lab and into space? Maybe lightweight transfer learning can help.
What are the major technological gaps to be filled in your group’s topic area? Locomotion, harsh environments, friction and environment models are difficult.
Which “magic wands” are still being used? Classification of the environment through hard-coding, not using perception.
Where does autonomy fit in with humans? It should replace them as much as possible---but the sheer lack of understanding we have of the challenges in space result in a need for in-the-loop problem solving, which only humans can do.
20sec delays are deadly for human teleop, nearly impossible to resolve without high-level planning interfaces and low-level autonomy.
“More risk taking would help”---would lead to new applications, rather than being tied to the ground.
NASA is engineering-oriented but science-driven; expense realities make it difficult to accept failure.

3 of 4

Space Robotics Competition

SRC and Valkyrie was a great testbed, presents challenges. New designs as parts of that platform were particularly challenging and limited---needs many iterations, e.g. in knee and elbow joints. Was a first step in difficulty.

SRC may evolve to something more, but unclear if actual problems were solved for space robotics.

Most difficulty was a result of the bandwidth restrictive, time-delayed systems, presenting the need to enable high-level teleoperation.

Modular components in design of space robotics very cool. When do we need that versatility? Can it be as reliable as customized platforms?

4 of 4

Take Away Ideas and Questions

Conservatism - system complexity = risk

Computational capabilities - how much is enough?

Funding/people

Confidence and trust with robot interaction

1 of 4

2 of 4

3 of 4

4 of 4