Strategy
Over the years, many people have commented that my sumo robots' line sensors must not be working very well. To which, I normally reply that the robot doesn’t have any line sensors. That comment results in looks of puzzled surprise. The fact is that sensing the line is not part of my sumo strategy, nor has it ever been! All of my sumo robots have always used the same, simple strategy. Step 1: find the opponent. Step 2: push.
Finding the opponent is dependent on multiple start patterns, good object detection sensors, and sufficient speed to attack before the opponent has a chance to escape or take evasive action. Push is dependent upon maintaining contact with the other robot, keeping the other robot centered directly in front, and reacting intelligently to the opponent’s attempts to escape.
Start Routines
The
goal of the start routines is to make contact with the opponent
immediately after the start. Preferably, that contact is made in
an advantageous location such as pushing on the opponent’s rear or side
rather that the opponent’s front. The local rules for starting
placement and knowledge of the opponents initial actions determine the start routine
utilized. NEIRG requires that the robots start side by side
on the start lines, facing in opposite directions. For that starting
configuration, there are four routines: (1) Search R/L, (2) Run Away,
(3) Back Away, and (4) Loop. For the CIRC event, the robots can be
placed anywhere behind the start line. For that starting configuration,
there are four more routines: (5) Hit Spin, (6) Direct, (7) Angle 45,
and (8) Angle 90. Routines 5-8 all begin with ExSpurt directly on the
start line. Each of the eight routines can be run to the right or
left.
Routine Videos
To give a better idea of the start routines, videos of ExSpurt running each routine are given below. The red LED flashes from 1 to 8 times to indicate the routine being initiated. The Right hand push button starts the selected routine to the left and the left push button starts the routine to the right. Actually, the right push button is the BS2SX reset and a brief tap the left push button increments the start routine by one. The number of flashes for each routine is included in the videos. Also, to save some download time, the start delay has been reduced from five seconds to one second. Each start routine is dead reckoning and is designed to end up in the center of the ring running in a searching loop. If another robot is detected at any time during any start routine, the start routine is terminated and the first attack program is initiated. In the videos, once the start routine has completed, the remote kill switch is used to stop ExSpurt.
1 Left Search R/L (0:08 - 0.8Mb) 1 Right Search R/L (0:08 - 0.8Mb)
2 Left Run Away (0:08 - 0.8Mb) 2 Right Run Away (0:08 - 0.8Mb)
3 Left Back Away (0:08 - 0.8MB) 3 Right Back Away (0:09 - 0.8Mb)
4 Left Loop (0:09 - 0.8Mb) 4 Right Loop (0:08 - 0.8Mb)
5 Left Hit Spin (0:11 - 0.8Mb) 5 Right Hit Spin (0:11 - 0.8Mb)
6 Left Direct (0:11 - 0.8Mb) 6 Right Direct (0:09 - 0.8Mb)
7 Left Angle 45 (0:12 - 0.8Mb) 7 Right Angle 45 (0:11 - 0.8Mb)
8 Left Angle 90 (0:12 - 0.8Mb) 8 Right Angle 90 (0:13 - 0.8Mb)
Sensors
For the opponent sensors, each of the five QEC113 IR LEDs
sends out a brief (0.4 msec 38 KHz) pulse. If that pulse reflects off
of an object, the reflection is detected by one or more of the four PNA4602M IR
receivers. Interrogating the five sensors takes only 2.1 msec. Each IR
LED covers 15° for a total of 75° horizontal coverage for the five IR
LEDs. The IR LEDs are aligned vertically such that they do not detect
the surface of the Dohyo, but will detect a pencil laying on the
surface. Since local lighting conditions vary substantially and the
PNA4602M sensors have built in automatic sensitivity control, the power
to the IR LEDs is adjustable over a very wide range. The sensors are
normally adjusted to detect the opponent at four inches, longer if there
are no outside objects near the sumo ring. Before the start while the red LED is pulsing
the number of the start routine, there is a one second delay between one
series of red pulses and the next. During that one second delay, the LED
indicates the condition of the sensors. If none of the sensors detect
anything, the LED is off. If one or more sensors detect an object, the
LED is green. If all of the sensors detect an object, the LED is red.
The LED display is actually updated 50 times during the one second
interval. Below is a video showing the LED display for each of the above
conditions. In the video the red LED is flashing once to indicated
that the Search R/L start routine is selected. Note that ExSpurt does not detect Mr. Lego at 8" or 7"
distances. At 6" the green LED flashes occasionally for a
partial contact. At 5"
the green LED is on constantly. At this distance, ExSpurt has a lock
on Mr. Lego. At 2" the green LED is still on solidly. At
0.5" the LED is red indicating that all of the five sensors on ExSpurt
see Mr. Lego.
Speed
ExSpurt presently uses a top speed of about 48 inches per second. At that rate, it can make contact with the opponent in about 0.5 seconds after the start. What that means is that the opponent must take the correct evasive action immediately at the start. If the opponent has any drop down or flip routines to get in proper configuration, it probably won’t have time to move before contact is made. So, assuming that the opponent takes evasive action at the start, which way should it go? The attack could come from most any direction. So it becomes a guessing game between the two robot operators. Will the opponent do the same routine twice in a row or pick an alternative. Could we call this a chess match? Since most present sumo robots have only one starting alternative, picking the best attack is simple.
Another positive result of high speed is that the chance of being attacked from the side or rear is very remote. For that reason, no side or rear sensors are provided. There will come a time when someone will build a faster and/or smarter sumo robot and I will have to modify my strategy. For now, 48 inches per second is fast enough but I do plan on experimenting with faster speeds in the near future.
Search
The underlying concept in push is that any direction that ExSpurt is pushing its opponent is toward the edge of the ring. Since any direction is good, no attempt is made to push the opponent toward the nearest edge. There are two programs that control push. The first program (Search) is designed to make initial contact with the opponent. After the start routine has terminated and no opponent has been detected, ExSpurt begins a search pattern in the middle of the ring. The search pattern is a simple loop. The speed of the loop is slow enough to spot the opponent entering the middle of the ring from the side, yet fast enough to prevent an opponent from catching ExSpurt's side or rear. Once an opponent is spotted, ExSpurt starts the chase. ExSpurt is designed to make fast, sweeping moves to close the distance to the opponent as quickly as possible. If contact with the opponent is lost, the search is continued in the direction that the opponent was last seen.
Destroy
The second program (Destroy) is designed to keep the opponent directly in front of ExSpurt. Therefore, quick changes of direction are specified. With the outside IR LEDs on the edges of the bumper, any minor slippage in the opponent to the right or left is immediately corrected. The next video shows ExSpurt pushing a very light Mr. Lego. ExSpurt can be seen overcorrecting each misalignment. With a typical (heavier and higher traction) opponent, the overcorrection disappears and the push looks much more linear.
Evasion
In the Destroy program, if contact with the opponent is lost, ExSpurt will assume that the opponent has spun to one side and is attacking from that side (that is a scenario that has actually happened in competitive matches). At that point, a backing maneuver is initiated to reacquire the opponent and restart the attack. The next video shows ExSpurt attacking a moving block, acquiring the target, losing contact, taking evasive action, and jumping between the Search and Destroy programs. This is how ExSpurt gets its exercise.
Problems
The
final problem scenario is a head to head pushing match. If one of
my sumo robots is to loose a match, this is how it happens. For
years I have experimented with optical mice in Extrasensory and
Executioner to identify this problem scenario. So far, in actual matches, the optical mice
have not detected sufficient rear movement to trigger evasive action.
See the match
4B Cheeky-san vs Executioner for a good example.
Presently, I am experimenting with a timing function to detect the
problem scenario. It will take a substantial number of future matches to determine if
the timing function proves beneficial.
Summary
Nowhere in any of the above is there a need for edge detection sensors. Even great amounts of pushing power are not needed. What is needed is a great amount of speed to implement the start routines. The opponent must be engaged quickly and be defeated quickly before he has time to react. Early Brooksbots with molded urethane tires were assumed to be great pushers and later Brooksbots with kill switches were assumed to be remotely controlled, but neither is assumption is correct. Actually, the intelligence of the software and strategy has won most matches. Everything else is intended to be a diversion.