I know a bit about robot's (I program them for a living) and I have seen a few very cool robots online including one that shot pool. I have put a cue on the end of a Fanuc 420 myself and shot a few shots. I think there would be a lot of work or investment needed to make a runout robot. More than anyone would ever spend since we cannot even get a decent sponsor for real pool. The primary issue's are vision, including calibration, decision making, interference (balls and rail). A robot would have to deal with all of that programmatically.
The way an industrial robot is programmed in most applications you teach snippets of motion and then you rerun those snippets as a part comes into the area, I have done lots of those and they do not include vision everything is fixed. I have also done projects using cameras to guide the robots this includes parts stopped take picture, adjust for part location and run motion relative to that part. I have picked up parts on moving conveyors linearly and rotating on dials so picked them up or put them down while the target is moving all of this fairly straightforward just lots of work depending on the actual application. However most all of these static in that we want to do the same thing repeatedly for a varying part location initial or destination or both. Applications are also done to tweak a part such as check it and then bend it or tweak it so the robot acts variably to normalize.
Another thing is that robots are very repeatable, they are not super accurate. This means they do not have a linear error over their entire workspace. This has caused issues to more than a few industrial projects. They usally linkages based on arms with linear length and these links are one attached to another. This can cause very slight error but error which stacks and becomes large. Pool needs to be played with a sub mm accuracy, robots are repeatable over their works space sub mm but are not accurate. Most fanuc robots are repeatable .08mm meaning they can go to a point in space move some where else and come back to that point with that tolerance, this a small number. However if you told a robot go a point X at one extreme locaiton in its reach and then told it to go to Y a calulated position at the far end of its reach and those 2 positons are calulated to be 2 or 3 meaters away and in extremly different orientation and the tip of the robot was measured at each location the error would be much more than .08 mm. Robots are repeatable not highly accurate. This can be improved with lasters and tuning a specific robot. But robots are not super rigid and there error compounds through there linkages as opposed to a CNC with very rigid axis composed primarily of directions 90 degrees oppsosed to each other. Kinematically and mechanically a very accurate sytem. The Deep Green robot below which is a gantry, appears to be a 5 axis gantry and is great for a lab a super high end pool robot would have to be much more rigid.
Projects I have never tried to do but that have been basically impossible till recently include shaft insertion into splines and bin picking. Shaft insertion includes placing a splined shaft into a hole such as a transmission and automatically aligning the shaft with external splines into a hole with internal splines. This is not a good appliaction for vision and via good application development it has been accomplished in the last few years using force feedback modules. This is still not simple however a person can pick a shaft up direct it into a hole and wigge it a few times and push it in and viola this is accomplished, not so easy for a robot but doable.
The other application is bin picking take a big deep box and fill it with parts to the top, the parts need to be picked up and loaded into a machine, and the parts need to be picked up in a certain orientation. This is being accomplished more and more but the issue of wall interference the decision of which one to pick or which area to look in has made this a very tough nut to crack. it is being accomplished more and more. I bet this is the result of 10 million plus in development by Fanuc alone there are other robot companies and they do not share info but they do demo breakthrus, this bin picking is becoming more commmon was once impossible.
http://www.youtube.com/watch?v=tsv6TLPpf3M
These two simple tasks to teach a human have taken a ton of research to accomplish and are now possible. This has been millions of dollars and thousands and thousands of hours from multiple people using a huge range of skills to accomplish these tasks, I believe a runout robot would take that and more. It could be done and is technically feasable, But without the same sort of funding and technical development as the Deep Blue team from IBM when challenging Kasparov or the Jeapordy guys I do not see it happening. The harware along would be well north of $250,000 just for an invert mounted robot (robot upside down on a track) on a 25 foot rail would cost that much before you put on any tooling or cameras. I have done some work with invert mounted robots and the final cells are all north of $500,000 and are only doing simple functions like picking things up and putting them down, nothing complex like playing poool. I do not think we will be seeing a runout robot anytime soon meaning a robot to challenge Archer or Efren.
That being said here are some lab examples which will hopefully disprove me, still wating for a robot with an actual stroke.
http://www.youtube.com/watch?v=AENJxqR0g48
I love these laser lines where the system draws on the table 1:25 or so
http://www.youtube.com/watch?v=LrMrFHtT8tE
http://www.youtube.com/watch?v=fhMgQ290NB0
http://spectrum.ieee.org/automaton/...ot-plays-pool-throws-down-robot-pool-gauntlet
Sorry for the long winded reply.
