Hi together,
we have some questions about the finals.
a) Will the scores from the preliminaries be taken into account for the finals? Or are they only needed for qualification?
b) How will the preliminary dexterity score be computed? Sum of the normalized score per task (Omni inspection, linear inspection, shoring, …)? It would be great to have scoring sheets available soon.
c) We need to plan our lab time and schedule with others. Is there already a rough schedule for the finals? Something like which discipline will happen on which day would help us to reserve the needed lab times. E.g.
Thursday: Exploration and Inspection
Friday: Maneuvering and Dexterity
Saturday/Sunday: Dexterity
d) What will the finals look like? Should we expect a subset of the tasks in the preliminaries for dexterity?
e) How will the final be performed technically? A video call where we stream the 4 cameras shown in the perliminary videos?
Thanks.
Best,
Kevin
Hi Kevin,
a: I’ll have to doublecheck to make sure I give you the correct answer. Note that this year we are only running best-in-class so really the “finals” is a chance for the best teams to do more tasks in front of a live audience. Given that we don’t know how much overlap there will be in the tasks that teams perform, and team availabilities, it may need to be a call we make once we know who is in the “finals”.
b: Yes, best scores normalised to 100. The scoring sheet for dexterity is as per pages 70-72 of the slides. I believe that the combinatorics are per task (inspect, extract, strength, shoring) although I will need to check if linear and omni split out separately (although it shouldn’t affect what you’re optimising for anyway).
c: We have kept the finals schedule flexible because we want to work around the teams and their timezones, which we can’t do yet as we don’t know which teams are in the finals. If you think you’ll get into the finals, book some lab time when it’s convenient for you and we’ll work around you.
d: As we still have no idea what the spread is likely to be, and the tasks that the teams can perform, we have decided to keep this flexible so we can be as fair as we can given the circumstances. We’re open to constraints you have at your end that may influence this.
e: The plan is to have a live Zoom call, exactly how will depend on the logistics and bandwidth available to the teams (and if they’re in a region that blocks Zoom). My hope is to have several people at the team’s end logged into the Zoom call with their phones and each streaming a different view as that seems to be something that should be readily achievable by most teams.
Cheers!
Thanks, Raymond, that sounds reasonable to me.
In the Strength Test Form is a field for the weight. How is this accounted for in the scoring? As a multiplier?
Hi Kevin,
Completeness of the set of tasks (pick it up, put it down on one side, put it down on the other side, carry out, and into the crate) is the primary measure. We want to know, for each height and reach, what is the maximum weight at which you can perform the whole set of tasks.
For two robots that complete all tasks in the same heights and reaches, the weight becomes the tiebreaker.
Of course this leads to the question, what happens if one robot does fewer heights/reaches but does so with more weight, while another does more heights/reaches but does so with less weight. I’ll have to check and get back to you on that one.
Cheers!
Thanks for the feedback. I am wondering if the tiebreakers will really matter in the end.
In dexterity we have 8 different disciplines, I expect the teams to perform on average up to 15 variations of each task. I think it is highly unlikely that this will end up in a tie.
Optimizing for time and efficiency, if the weight only counts as a tiebreaker we would just use the weight of 1kg for all tasks.
I would suggest adding the weight as a multiplier, a linear multiplier is probably too strong. Somethings like multiplier(weight)= log_10(weight)+1 might work well if the minimum weight is 0.5kg.
The multiplier would be
0.5kg -> 0.7
1kg -> 1
5kg -> 1.7
10kg -> 2.0
I think this would reflect the increased difficulty for handling heavier weights while keeping the scores in the same ballpark.
Hi Kevin,
So the goal is to measure the robot’s full capabilities. If for a given height and reach you can do 2kg, there should be no incentive to do it at 1kg. Of course this presumes that if the weight is too heavy, you have some way to stop the test and re-do at a lighter weight without damaging the robot so please do take care in that regard!
Note that the weight can vary between different heights and reaches so we will find out at which ranges you can lift which weights. You can certainly choose 2kg at the closest reach, and 1kg or 0.5kg at the furthest.
We have a meeting today where I will try and clarify how we trade off total reach space and weight and get back to you as while that doesn’t affect the strategy for running the test itself, it may affect your choice of robot configuration.
Cheers!
Hi Kevin,
I’ve just posted a clarification thread (as a new thread so the subject is obvious). Please continue the discussion over there if you have any further questions!
Cheers!