PDA

View Full Version : Autonomous or Semi-Autonomous Roll Call



parallax
07-17-2010, 10:47 AM
Out of curiosity...

Who out there is thinking about competing in this years competition with an autonomous or semi-autonomous mech? :)

Furthermore, as much of a challenge that it is to make a competitive autonomous mech is, does it detract from or add to the spirit of the competition? Why or why not?

gdubb2
07-17-2010, 11:03 AM
I think a fully autonomous mech would be great. But extremely difficult to do, given the way the arena is laid out with buildings, and decor. However if someone was able to pull it off, I for one would be giving mucho kudos for the ability to do it...:veryhappy:

Gary

parallax
07-17-2010, 11:15 AM
@gdubb2: Can you elaborate on the difficulties you see presenting themselves based on the decor and arena design? Assuming the playing surface is flat again, I don't see how obstacle avoidance sensors and behaviors would have to be much more specialized than what we typically see (i.e. rangefinders, bumpsensors, etc). However I wasn't there last year, and am probably missing something...

parallax
07-17-2010, 11:24 AM
Sorry to double post, but I also want to make it clear I am not underestimating the challenge presented by creating and competing with an autonomous mech :-)

lnxfergy
07-17-2010, 11:37 AM
@gdubb2: Can you elaborate on the difficulties you see presenting themselves based on the decor and arena design? Assuming the playing surface is flat again, I don't see how obstacle avoidance sensors and behaviors would have to be much more specialized than what we typically see (i.e. rangefinders, bumpsensors, etc). However I wasn't there last year, and am probably missing something...

I imagine you could get a mech to navigate a bit autonomously -- it wouldn't be much tougher than the fire fighting competitions. However, actually finding your opponent and scoring a hit on them is an entirely different matter -- and I think this is where the difficulty comes in.

I've several times said/thought that a "fly-by-wire" system, where the mech can auto-correct it's course, would be very helpful (especially if your camera is dropping in and out, you might be able to run away from a fight while it settles down). However, I'm not entirely sure how well auto-targeting is going to work. MissAlignment was using some sort of auto-targeting by placing pink postIt notes on the opponents plates -- but I don't recall seeing how well it was working. The new building has much nicer lighting than the old one -- but still had some of the same issues of lots of sun washing out the arena late in the afternoon ("hey, who's shining a flashlight at Giger?"... "Andrew, that's the sun coming in through the windows").

-Fergs

gdubb2
07-17-2010, 11:55 AM
Parallax, Fergs pretty well covered it. I'm sort of building a fire fighting bot, and just experiencing the problems with it, I can't imagine adding in targeting, firing, and avoiding getting shot. I guess a better way to put it, is that this wouldn't fall within my very limited skill set.

Gary

parallax
07-17-2010, 12:27 PM
I see your points!

The technical challenges posed by a fully autonomous mech given the variable lighting conditions, unpredictable arena, and other considerations such as r/f interference (for any off-board processing) are considerable to say the least. Knowing that there are people on this forum that have far more skill, experience and education than myself are not attempting it because of its difficulty goes to show me just how much of a challenge it really is!

Also I see that it has been a point of hot debate in the past. After I started this thread I began diving into the related thread called "on encouraging autonomous mechs." If you read through that it becomes apparent that people have strong feelings on the issue either way.

I know that this past year some attempts were made for semi-autonomy (such as MissAlignment's autotargeting feature) and I am sure we will see some more "pilot assist" features this year as well. Sounds like you might try out something along those lines, Fergs? I myself have been slowly piecing together some systems to "pilot assist" with my mech. Of course, that will all be dependent on whether I can make it out to RoboGames at all, given the travel expense+build costs.

I guess what I was trying to do by starting this thread is see who else out there is trying something like this- either a semi-autonomous or fully autonomous mech, and what kinds of things they are doing with them.

I am reminded of the DARPA grand challenge in a way. The first year no one made it out of the starting gate except for CMU's "Sandstorm" which ended up getting caught up on a rock and setting it's own tires on fire trying to get off it. The next year was a different story, however... As the teams tackled the issues that set them back the year before and learned from the other competitors ideas, the second year had several vehicles finish.

The first year of Mechwarfare showed no autonomy at all. Last year we saw early attempts. I wonder what we will see this year?

sthmck
07-17-2010, 01:17 PM
The first year of Mechwarfare showed no autonomy at all. Last year we saw early attempts. I wonder what we will see this year?

I don't think we will see much more in the way of autonomy from last year to the next. I think most of the people that built mech realize one of two things.
1: It is hard enough to build a functioning mech that can be competitive let alone one that will actually function.
2: Those who have built mechs capable of being competitive realize that there is probably a lot more that they can do with their current designs.

Reliability takes precedence over autonomy, just as maneuverability took precedence over fire power in the previous two years. (with a few exceptions)

I came to robogames with what was essentially a partially completed mech and I still managed to place fairly well all things considered. While I would love to have semi autonomous system in place on my mech, I am aware of the fact that I can become much better just by improving on my overall design and making my mech more reliable.

I think effective autonomous systems are still a couple years off. If you look at the actual number of mechs that could complete a full match without any problems at all compared to the number that entered you will see that there is still work to be done by builders before we can have effective enough platforms capable of allowing autonomous function.

You have to crawl before you can walk and most of us are still crawling.

Upgrayd
07-17-2010, 03:23 PM
Can you elaborate on the difficulties you see presenting themselves based on the decor and arena design? Assuming the playing surface is flat again, I don't see how obstacle avoidance sensors and behaviors would have to be much more specialized than what we typically see (i.e. rangefinders, bumpsensors, etc). However I wasn't there last year, and am probably missing something...

Along one wall cop cars were lined up narrowing the path. Buildings had cars sticking out of them in random areas. All of the buildings were surrounded with about a 1/8 inch tall sidewalk. Sidewalks had scale benches, street lights, ect.

All of these things were very easy to get caught on.

lnxfergy
07-18-2010, 01:47 AM
Also I see that it has been a point of hot debate in the past. After I started this thread I began diving into the related thread called "on encouraging autonomous mechs." If you read through that it becomes apparent that people have strong feelings on the issue either way.

I personally don't think that the autonomy is the hot debate -- several previous threads were bogged down by requests to modify the game so as to make autonomy easier or more relevant. As long as you don't ask for modifications to the arena or game play, I don't think there's much contention. Note however, that we've added a qualification requirement to MW2011 -- this will also take care of the second half of the argument, because a fully autonomous bot would have to prove itself before capable before being put into the competition ladder. As I typed that last sentence, I realized we probably have to clarify the qualification rule to state that fully autonomous mechs must run the qualification autonomously (or more generally, all mechs must run the qualification as they will run regular matches).


I know that this past year some attempts were made for semi-autonomy (such as MissAlignment's autotargeting feature) and I am sure we will see some more "pilot assist" features this year as well. Sounds like you might try out something along those lines, Fergs?

If I ever find enough time, I'll have a mech with onboard obstacle avoidance and a fly-by-wire control to guide it. A fly-by-wire system would be a huge asset (as long as it's very reliable and intuitive, if the pilot is fighting the system, there will be no gain). Upgrayd brought up the cop cars, I had actually forgotten about them, but the sidewalk and other obstacles out there were very hard to see (especially with wide quads, where the camera FOV doesn't show what your leg may be hitting). Several bipeds tripped over the sidewalks, quads got hung on buildings (sometimes causing hits on their own plates), etc. A robot that can guide the operator around obstacles outside of the camera FOV would be a huge asset. Let's face it -- even with piles of computing power available to bigger robots, we humans are still better at quickly analyzing a scene (and finding the opponent mech, and the general direction we want to go). Studying fly-by-wire and UAV technology would be a good starting place -- they've done lots of work to let pilots still set the big picture goals, while the machine handles things like stability (or, in our case, obstacle avoidance, and possibly targeting if you can get it to work). This is where semi-autonomy gets the most bang for it's buck, in an augmented system.


I am reminded of the DARPA grand challenge in a way. The first year no one made it out of the starting gate except for CMU's "Sandstorm" which ended up getting caught up on a rock and setting it's own tires on fire trying to get off it. The next year was a different story, however... As the teams tackled the issues that set them back the year before and learned from the other competitors ideas, the second year had several vehicles finish.

The first year of Mechwarfare showed no autonomy at all. Last year we saw early attempts. I wonder what we will see this year?

They also threw a HUGE pile of money and people and other resources at the problem. I wouldn't expect a working fully autonomous mech that is even approaching on-par with the current mid-pack bots for several years at least. While I'm sure you could probably build a bot that qualifies (maybe even just dead reckoning it), there's a big difference between a bot that is capable of competing, and one that is competitive.


I came to robogames with what was essentially a partially completed mech and I still managed to place fairly well all things considered. While I would love to have semi autonomous system in place on my mech, I am aware of the fact that I can become much better just by improving on my overall design and making my mech more reliable.

Friday qualifications are going to pretty much rule out the "show up with a partially completed mech" - I know Seth knows this, just wanted it to be out there for everyone else. Have your mech ready to compete when you arrive.

Finally, reliability is the most important thing with this competition -- don't underestimate just how much can go wrong. Even with 1 timeout that each competitor was allowed throughout the weekend (note, this will not be the case in 2011, no timeouts) I think we had only 1 bot that didn't default a round due to technical failure. Guess which one? The one that won the competition.

-Fergs

parallax
07-18-2010, 10:15 AM
They also threw a HUGE pile of money and people and other resources at the problem. I wouldn't expect a working fully autonomous mech that is even approaching on-par with the current mid-pack bots for several years at least. While I'm sure you could probably build a bot that qualifies (maybe even just dead reckoning it), there's a big difference between a bot that is capable of competing, and one that is competitive.


Absolutely! I agree 100 percent! If I remember correctly, CMU had a several million dollar warchest and a team of 60+ people. I wasn't attempting to suggest that we would see improvements in mech autonomy/technology on par with what happened at the DARPA Grand challenge; I was just highlighting the improvements from year to year that tend to manifest themselves in competitions which have the same overall goal. In my (albeit limited) experience, this tends to be the case.

Like everyone has pointed out, autonomous mechs are probably several years out and they'll probably just be at the point where they will compete; not even necessarily be competitive. I like that people are thinking that it is something we'll see in the future though.

I see what you are saying about UAV technology Fergs, and it makes total sense. I also hope that you are able to find the time in between now and RoboGames to implement some of your ideas :) My thoughts for a mech were similar, to create a "pilot assist" and therefore some level of autonomy.

I was focusing on targeting assistance rather than navigation for my hypothetical systems. I understand the difficulties in place for such a system, but I have a soft spot in my heart for machine vision and image analysis. In fact, I'm sad to say that my hex Ally is far better at tracking faces than walking at the moment! However I'm looking forward to trying to apply some of lessons I learned with Ally into my mech.

Thank you all for sharing your thoughts on this. I think this thread is a good indicator of any level of autonomy we will see at this years competition, and has shown that it is more of a future goal than a present reality

lnxfergy
07-18-2010, 11:41 AM
I was focusing on targeting assistance rather than navigation for my hypothetical systems. I understand the difficulties in place for such a system, but I have a soft spot in my heart for machine vision and image analysis. In fact, I'm sad to say that my hex Ally is far better at tracking faces than walking at the moment! However I'm looking forward to trying to apply some of lessons I learned with Ally into my mech.

Yeah, Issy2 was easily capable of a 50-60cm/s run, however, I couldn't drive it that fast through the camera view. Even though it walks very straight, if I had the slightest angle in my heading, I'd be in the wall before I could really correct it. I had him tuned down to about 20cm/s which was more manageable.

-Fergs

RobotAtlas
07-22-2010, 02:25 PM
I think this thread is a good indicator of any level of autonomy we will see at this years competition, and has shown that it is more of a future goal than a present reality

It is going to be a future goal for a long time if we keep working separately, duplicating effort, working on things that have been already worked on (servo reselection, controller redesign, etc).

The only way to speed things up is to come up with well defined and achievable goals.
Since we are all hobbyists here, we have to use our time as efficiently as we can.

lnxfergy
07-22-2010, 07:40 PM
It is going to be a future goal for a long time if we keep working separately, duplicating effort, working on things that have been already worked on (servo reselection, controller redesign, etc).

The only way to speed things up is to come up with well defined and achievable goals.
Since we are all hobbyists here, we have to use our time as efficiently as we can.

I'm actually more skeptical of autonomous mechs because of who is currently participating. I think that autonomy is also not exactly high on most people's list of things to do -- most of our current competitors are much more mechanically oriented than they are towards programming. Just take a look around at the events at RoboGames in general, only a handful of bots are autonomous out of hundreds there. Autonomy is mostly found in the Fire Fighting and RoboMagellan contests -- where the goals are a bit more well defined (especially in terms of what sensors really are available at low cost, and the processing power required to handle them). Even with the better defined goals, none of the bots in RoboMagellan this year completed the course, and a majority of Fire Fighting runs ended in failure.

I think we've also got a bit more work to do just on the main competition:


Quad walkers have come a long way, they're mechanically stable, but bipeds still need much better gaits.
We've found fairly reliable control using XBEEs, but there's still some work left there to improve how well a bot can be driven.
The video feed caused quite a few problems both years. RF interference just sucks.

Even with all the information that has been shared around, it really will take most people a year of actually doing this to realize exactly what needs to be done.

-Fergs

RobotAtlas
07-23-2010, 06:17 PM
We've found fairly reliable control using XBEEs, but there's still some work left there to improve how well a bot can be driven.


Which reminds me to ask how come XBEE was dropped in Issy3?
Not enough COM ports, right?
But then wouln't we have communication problems?

lnxfergy
07-23-2010, 07:16 PM
Which reminds me to ask how come XBEE was dropped in Issy3?
Not enough COM ports, right?
But then wouln't we have communication problems?

The ArbotiX is connected with a hardwire USB to the FitPC, and then I use SSH, remote desktop, or my usb keyboard/mouse and a monitor to access the FitPC. I wouldn't recommend wifi for mech warfare, but Issy3 isn't going to be participating in mech wars.

-Fergs

RobotAtlas
07-23-2010, 08:03 PM
I wouldn't recommend wifi for mech warfare, but Issy3 isn't going to be participating in mech wars.

-Fergs

Why not? You don't want to win? :)
When other robots at the next RoboGames have assisted piloting, whouldn't you want one too?
Combine that with rather problematic solution of using a WiFi camera in Robo-Games, and there's a need for a better comminication.

The top speed of XBee Pro of 250k (according to this: http://www.digi.com/products/wireless/point-multipoint/xbee-series1-module.jsp#overview)
is not going to do video, unless video compression is involed, but then we are back to FitPC2 to do compression, aren't we?

Video over XBee - have anybody tried it?

darkback2
07-23-2010, 08:17 PM
I was probably the only person using wifi for control last year, and lost one match because I lost the connection. I would suggest in the least bringing a bluetooth backup. Also, Connor was managing the network and participating in mech warfare. He set up a dedicated network for each match, which means if your controlling your robot over wifi alone, you will only be able to control it when your in the arena having a match. Next year I hope to bring another solution so that I can test my bot outside of the arena.

Xbee seamed a lot more stable. People stored gates, or used nuke on their bots, and then were able to control their robots by sending rather small messages. like go forward. while the same is true for wifi, without auto recovery and stuff you can get into a bit of trouble.

Hope this helps.

DB

lnxfergy
07-23-2010, 08:43 PM
Why not? You don't want to win? :)
When other robots at the next RoboGames have assisted piloting, whouldn't you want one too?
Combine that with rather problematic solution of using a WiFi camera in Robo-Games, and there's a need for a better comminication.

Issy3 is a very different machine than a Mech -- he's going to be a bit slower and more methodical. I really don't see the need for a PC on a mech -- a skilled programmer can get assisted driving using a micro + some sensory (and it's likely to be more effective than a visual solution which needs a PC).



The top speed of XBee Pro of 250k (according to this: http://www.digi.com/products/wireless/point-multipoint/xbee-series1-module.jsp#overview)
is not going to do video, unless video compression is involed, but then we are back to FitPC2 to do compression, aren't we?

Video over XBee - have anybody tried it?

You're really not going to do video over XBEE, I doubt you'll see reliable transmission at the 250k mark at RG. 115k may work, but of course you're really saturating the bandwidth (which means you don't have much time for packet resends, which means packets just get dropped if they don't go through the first time). If you want any sort of decent frame rate you'd need compression technology that really doesn't exist today (and probably have very high lag).

-Fergs

DresnerRobotics
07-24-2010, 08:47 AM
Also, given the option of holding Mech Warfare in its own wall-separated room next year, I doubt we're going to have nearly as many WiFi issues. The only issues we had with Xbee was that the PC dongle wasn't close enough to the arena at first- we're mounting it up on the ceiling of the arena this year.