CASE 16 Workshop on Multi-Robot Systems in Automation: Topics in Planning and Control. Stocktaking. Dylan A. Shell TAMU

Transcription

1 CASE 16 Workshop on Multi-Robot Systems in Automation: Topics in Planning and Control Stocktaking Dylan A. Shell TAMU

2

3 Stock? lt/files/styles/420x420/public/styles/m ain/public/chicken-stock-xl.jpg

4

5

6

7

8 Why have we had so little impact on multi-robot systems in industry?

9 Why so little impact on multi-robot systems in industry? Timing: too early? Maybe if we just wait, we ll see more of the sorts of techniques we ve been developing actually used?

10 Why so little impact on multi-robot systems in industry? Timing: too early? Maybe if we just wait, we ll see more of the sorts of techniques we ve been developing actually used?

11 Why so little impact on multi-robot systems in industry? Maybe we ve got some things wrong? Studying the wrong or ill-conceived problems? Wrong systems? Maybe we have the wrong values?

12 Why so little impact on multi-robot systems in industry? Maybe we ve got some things wrong? Studying the wrong or ill-conceived problems? Wrong systems? Maybe we have the wrong values?

13 Consider the future with driverless cars... They ll be large-scale systems Autonomous robots, local decision-making Compare to the MRS in most of our papers...

14 Consider the future with driverless cars... Queuing considerations are important. (Goals/Tasks come from a variety of tasks) They ll be large-scale systems Deliberation mainly at level of picking non-congested routes. Some re-balancing for ride-sharing. Autonomous robots, local decision-making Compare to the MRS in most of our papers...

15 How about warehouse systems? They are large-scale systems Compare to the MRS in most of our papers...

16 How about warehouse systems? They are large-scale systems Compare to the MRS in most of our papers... We don t really study the logistics of continually running systems (bringing some robots up, some down, churn ) with maintenance and monitoring anywhere near enough. We seldom consider the problem of designing the robots and the environment together; the objective is to jointly optimize over all the infrastructure.

17 Long-term deployments mean robots malfunction; aka personalities Too little impactful work on self-monitoring; learning recurring glitches, etc. How about warehouse systems? We ve studied heterogeneity but the wrong sorts. They are large-scale systems Compare to the MRS in most of our papers... We don t really study the logistics of continually running systems (bringing some robots up, some down, churn ) with maintenance and monitoring anywhere near enough. We seldom consider the problem of designing the robots and the environment together; the objective is to jointly optimize over all the infrastructure.

18 How about systems?

19 How about systems? The purpose of systems is different in academic research. Yet, we don t have science of systems for robotics.

20 How about systems? The purpose of systems is different in academic research. I m pragmatic about what such a thing might look like: What does a system paper teach? Where are the generalizable lessons? What ought it to teach?

21 The systems in our labs are general platforms; industrial solutions fit a niche How about systems? The purpose of systems is different in academic research. I m pragmatic about what such a thing might look like: What does a system paper teach? Where are the generalizable lessons? What ought it to teach?

22 Wrong values? We generally want formalism, guarantees, optimality on crisp abstractions. We believe in the idea that there is a super-structure so we can study MRS topics.

23 Wrong values? We generally Increasingly want Common formalism, Criticism: guarantees, optimality on crisp abstractions. All this work represents merely point solutions We believe in the idea that there is a super-structure so we can study MRS topics.

24 Right values

25 Right values Nothing is more practical than good theory - Einstein But burgeoning industry will change the role of universities.