Deep Green - A Pool Playing Robot?

o0zi writes "A Canadian scientist has created another game-playing machine, designed for a far simpler purpose than chess: playing pool. The world's first pool-playing robot consists of a slim box that glides along tracks above a pool table, and shoots using a camera-guided cue. Deep Green pots only half the shots it plans for - supposedly the same as a below average player - but this is expected to improve."

Horizon snooker robot

[rjforster]

Any British /.ers remember a Horizon[1] episode where they built a snooker playing robot. Must have been 10/15 years ago now. Played on a reduced size table with fewer balls (10 rather than 15 reds IIRC). The gantry for the robo-cue included steel pillars at the corners of the table, thus making it really hard for the human competitor.

[1] Horizon is a science program on BBC2.

Simpler eh?

[Rosco P. Coltrane]

designed for a far simpler purpose than chess: playing pool.

This comment shows the poster has no idea what playing pool is about.

It's more than just line up / aim at the center of the ball / shoot more or less hard : you have to pot the ball, yes, but you also have to replace your white ball so that the next shot is easier. Often you have to think 2, 3, 4 shots ahead. Often you plan your entire game before playing the first shot.

In order to control the white ball, there's a certain about of spin to give it on the vertical plan and horizontal plan (english) so that the ball is deflected differently on the cushion(s), depending on the angle they arrive. Giving english to a ball also deflect its path (it won't roll straight), so that has to be accounted for in the aiming (you aim a little off). And then all tables don't react the same, some have newer, less "grabby" cloths than others... Then there's the roughness of the cue tie and the chalk, and the suppleness of the cue's wood that affects greatly how much english is put on the ball. Then of course there are all the "special" shots, like massés, that require a lot of practive to control... etc...

Playing pool is a LOT more complex than chess, and that's not just because it involves real physics. The problem has many many variables, and it takes many years of practice to master. I've been playing for 20 years, at least 2 hours per day, and I still couldn't beat a professional. It's a very demanding game.

This is a great invention

[maxeypad]

This has been done before to some degree with a robot called Iron Willie [allsportsbid.com] by a company called Predator whom used the robot to create low deflection cues and empirically measure how "accurate" cue designs were. Predator Cues [predatorcues.com] are to pool what high dollar putters are to golf. These cues utilize a pie wedge design in the shaft combined with a stiff taper and lightweight, short ferrule to decrease deflection and maximize energy transfer to the cueball. Many people report a 10-20 percent improvement when they start using a predator shaft on their cue. In fact, more professional players choose to play with predator shafts without sponsorship than any other cue on the market. The truly exciting thing about this invention is the fact that it will be used to create better pool equipment for independent testing. this robot will be much more flexible than iron willie and will be able to measure the performance of cues on a much lower level I'd imagine. Plus, i imagine the robot to be much more flexible than iron willie who simlpy can be setup to shoot the same shot over and over. Pool is such a complex game that it will be very difficult for the robot to get such concepts as sacrifice safeties and intentional fouls. Other games like one pocket, the pool equivalent of chess, will be really difficult to grasp for a computer since its very common to "sacrifice" a ball to your opponent for the good of the game. Read more about one pocket here [onepocket.org]

Re:Simpler eh?

[Miniluv]

Even without mastering spin and such like given a high enough degree of accuracy it is likely that a very good execution plan could be calculated.

Unless spin and the like are mastered a great many execution plans are removed from the realm of possibility. There are fundamentally two things that separate pro level players from amateurs: Planning and cue ball control. Planning is a lot easier to learn than cue ball control, since you simply have to know whats possible. Executing that plan requires a very, very subtle blend of speed control, english, and cut angle.

Rather than searching forward like in chess a pool playing robot will have the luxury to decide which ball it wants to pot and then extrpolate back using simple newtonian mechanics.

This isn't entirely true. Yes, in 8 ball you know that the 8 ball is your final destination, however there are usually several paths there. In fact, the better the robot is at cue ball control, the more possible paths and the higher percentage they all become. Even worse are games like 3 cushion carom, or snooker, where the goals are less definable. Sure, in snooker the robot knows the final goal is to sink the black ball...but there're 15 reds with a required color in between, then 7 colored.

Furthermore, despite the stated goal of sinking any ball in any pocket from anywhere on the table, this just isn't feasible. Every game you see situations where certain balls are only possible in certain pockets, and sometimes this means the only shot available is a safety. In snooker this is particularly true, since a well played safety can score quite a few points with forced continues. Try teaching the robot to properly judge safety shots, since they require a decent understanding of whats possible, and the robot just can't use its own skills as a baseline.

Another piece of prior art

[Anonymous Coward]

They didn't do much search for prior art, did they? Here's another billiards robot [autonlab.org], put together more than ten years ago by a guy who's currently a professor at CMU [cmu.edu]. (Look at the last page of the paper for a very grainy photo of the robot.)

The cool thing about this robot is that it learns from experience: it watches to see where the ball goes, learns a model of how that depends on its stroke parameters, and tries to compute a better way to sink it next time. As pointed out by another poster, it doesn't plan ahead to the next shot, which is an important aspect of the game.

Re:What's really needed is...

[iantri]

I know this was a joke.. but they really do exist.

In fact, there is one cleaning my pool right now. Depending on the design, it may or may not work very well -- mine uses the suction of the pump system to generate a "jerking motion", which moves the vacuum around the bottom. It, however, tends to go in predictable patterns (moving the hose around helps a bit) and stirs up a lot of the dirt before sucking it up.

Mine is similar to this model. [epinions.com]