Good point, I meant android but better safe to be sorry.
Originally Posted by W. Rabbit
I know the machine in Dune I mean, well I think I do, there's at least two and neither run on AI, just feedback from a human, like video game does today. Wicked concept but modern mechanics probably aren't up to it.
Good luck though, I'd love to see a working prototype one day.
You and me both.
Originally Posted by Kovacs
Mechanics of the project would have been impossible ten years ago simply due to cost but turns out in the field of humanoid robots, things have come a long way (my guess: sex robots will drive this just like porn created the Internet. Thanks Real Doll!)
There is a lot of humanoid research being done in academia with Pneumatic Air Muscles (PAMs), which are lightweight, relatively cheap, and easy to use in robotic designs. They have a couple of characteristics that I think would work here, and for those same reasons they are used extensively in things like the Shadow Hand that require a "human like grip". I like these very much...
Couple of brainstorming tossouts in case anyone has time or interest:
- Need to gather data on the average reaction times are for agonist/antagonist muscles (human) when attacking or counterattacking in a live sparring situation.. In other words, what is the avg/min/max time (assuming millisecond ranges) from the decision to punch a target to the moment of impact, based on the characteristics of human arm muscles. If anyone has ever had their strikes actually metered/measured by actual equipment in this way, I'd be very interested in the numbers (there is plenty of stuff online...I am looking for more).
- Anyone familiar with the Sagittal, Coronal, and Transverse planes of the body?
- Am I using any terms/anatomy here incorrectly?
This got dropped in the beginning of the thread because of the time-stamp errors:
Originally Posted by Zendokan
Originally Posted by Jiujitsu77
Originally Posted by Humanzee
Originally Posted by jk55299 on Keysi Fighting Method
The real deadly:
I'll have to digest what you've said, not sure the processing volume would be nearly that difficult.
Originally Posted by Zendokan
There have been a lot of miniaturization and processing advances in the last few years, you can build embedded systems with dual core Pentiums and controller area networks that can fit the processing power of midrange servers onto a small onboard circuit.
There are already way more complex robotics designs out there in the cost/price range of consumers (I use the $10,000 mark as an upper bound on total materials cost...which would include onboard/offboard computing, chassis, actuators/PAMs...lucky for me there is a LOT of free GNU licensed software available for robotics CAD and IO. PVC and air muscles are pretty damn cheap by industrial standards. :)
Was curious how you came up with this 5 Hz sample rate?
proces so much data at a max cycle rate of 0.2 seconds constantly for a period of 5 minutes.
Also, in terms of the "adaptivity" of the designs I am working simultaneously on modeling 3 different (initial) prototype ideas, none of which will end up being "adaptive" just yet, since that can come in a later phase and would involve a lot more complexity in terms of code. For simplicity in initial designs most of these will be either controlled from an offboard box (I have a spare 8-bit Nintendo controller I think would work perfectly) or from a script (L R L R L R or something simple).
Type 1 prototype: Static punching/mobile target bot (think punching bag with compliant pneumatic extensor arms)
Type 2 prototype: Static swinging/rotating arm dummy bot (think wooden shaolin dummy with moveable/replaceable sections that move according to a program/script)
Type 3 prototype: Static noncompliant grappler bot (a bot a grappler would have to submit for a "win", essentially a ground version of Type 2).
Last edited by W. Rabbit; 11/12/2010 1:00pm at .
You could do a simple version of bot 1.(At least for boxing) Just have a motion sensor in the front and have it shoot out preset combinations. Maby a hinge at waist level. and a rotaty thing to track you as you try to move around it.
Type 2 is the most straightforward concept but seems like it might tend towards a conditioning device (any gong fu people who do da sam sing arm conditioning might really appreciate being able to do it without a partner or doing it a more "live" mode that is possible with a wooden dummy).
Originally Posted by gregaquaman
Type 1 is the easiest to visualize being more useful as an actual striking/avoidance tool (think a Rock Em Sock Em Robot vs Human, same concept). It is actually very easy to make a robot that "follows" a target in front of it (ie I move left, it rotates right to adjust facing) but it involves using photoreceptors which are expensive and easy to knock out the calibration on. Not impossible though, and probably well within cost. Other than turning, getting into computer vision would be neat (does opponent have hands up (DEFEND) or is there an opening (STRIKE!). Lots of complexity there.
Type 3 is easier than it sounds as long as I stay away from something that will try to submit a human. That seems like the long shot so far. It is much easier to build a bot that a human must submit...basically just a bot that would keep trying to right itself depending on where/how pressure was being exerted on it. With the right tuning possibilities I think this would make a great solo conditioning tool for judoka.
Interestingly enough my patent searches so far did come up with a "boxing robot" from 1985 which doesn't fit any of my descriptions I have, but US patents only have a 20 year lifespan making it a non-issue.
What about them specifically?
Originally Posted by W. Rabbit
There are a lot of factors that make this unfeasable at least right now. First off you have complexity. To realisitcally mimic the abilities of even a beginner boxer or grappler is a level of complexity that engineers have not been able to meet, if you can get a robot to pull an omoplata than you will get a free ride at any college you want.
That said its reasonably possible to make a scaled up rock-em sock-em robot but its not likely to be as reactive as a real human sparring partner. You could also in theory make a robot that resists or attempts to apply a basic lock but it would only really work in a specific situation and so would not be very useful as a training tool. Even this would be extremely difficult.
You also have the problem of wearing out components. Anyone who uses grappling dummies or heavy bags know that no matter how heavy duty they are over time they wear out. Now ad complex components and the life span will drop significantly. Imagine putting your laptop in a heavy bag and wailing on it.
I think we have all had the idea of a sparring bot at some point (usually when our sparring partner bailed on us for the 100th time) but the technology just isn't there yet.
Single Sign On provided by vBSSO