WEBVTT 00:00:00.480 --> 00:00:01.990 Surprise! 00:00:01.990 --> 00:00:07.060 We've just been given a robotic ant that has an FSM for its brain. 00:00:07.060 --> 00:00:12.879 The inputs to the FSM come from the ant's two antennae, labeled L and R. 00:00:12.879 --> 00:00:18.570 An antenna input is 1 if the antenna is touching something, otherwise its 0. 00:00:18.570 --> 00:00:22.279 The outputs of the FSM control the ant's motion. 00:00:22.279 --> 00:00:27.690 We can make it step forward by setting the F output to 1, and turn left or right by asserting 00:00:27.690 --> 00:00:31.689 the TL or TR outputs respectively. 00:00:31.689 --> 00:00:36.910 If the ant tries to both turn and step forward, the turn happens first. 00:00:36.910 --> 00:00:40.760 Note that the ant can turn when its antenna are touching something, but it can't move 00:00:40.760 --> 00:00:42.640 forward. 00:00:42.640 --> 00:00:47.109 We've been challenged to design an ant brain that will let it find its way out of a simple 00:00:47.109 --> 00:00:50.489 maze like the one shown here. 00:00:50.489 --> 00:00:56.680 We remember reading that if the maze doesn't have any unconnected walls (i.e.,no islands), 00:00:56.680 --> 00:01:01.649 we can escape using the "right hand rule" where we put our right hand on the wall and 00:01:01.649 --> 00:01:05.010 walk so that our hand stays on the wall. 00:01:05.010 --> 00:01:08.820 Let's try to implement this strategy. 00:01:08.820 --> 00:01:12.880 We'll assume that initially our ant is lost in space. 00:01:12.880 --> 00:01:17.760 The only sensible strategy to walk forward until we find a maze wall. 00:01:17.760 --> 00:01:23.420 So our initial state, labeled LOST, asserts the F output, causing the ant to move forward 00:01:23.420 --> 00:01:28.539 until at least one of the antennae touches something, i.e., at least one of the L or 00:01:28.539 --> 00:01:32.770 R inputs is a 1. 00:01:32.770 --> 00:01:37.270 So now the ant finds itself in one of these three situations. 00:01:37.270 --> 00:01:41.509 To implement the "right hand rule", the ant should turn left (counterclockwise) until 00:01:41.509 --> 00:01:45.070 it's antennae have just cleared the wall. 00:01:45.070 --> 00:01:50.560 To do this, we'll add a rotate-counterclockwise state, which asserts the turn-left output 00:01:50.560 --> 00:01:55.430 until both L and R are 0. 00:01:55.430 --> 00:02:00.010 Now the ant is standing with a wall to its right and we can start the process of following 00:02:00.010 --> 00:02:02.030 the wall with its right antenna. 00:02:02.030 --> 00:02:06.540 So we have the ant step forward and right, assuming that it will immediately touch the 00:02:06.540 --> 00:02:07.960 wall again. 00:02:07.960 --> 00:02:12.609 The WALL1 state asserts both the turn-right and forward outputs, then checks the right 00:02:12.609 --> 00:02:16.250 antenna to see what to do next. 00:02:16.250 --> 00:02:22.049 If the right antenna does touch, as expected, the ant turns left to free the antenna and 00:02:22.049 --> 00:02:24.190 then steps forward. 00:02:24.190 --> 00:02:29.950 The WALL2 state asserts both the turn-left and forward outputs, then checks the antennae. 00:02:29.950 --> 00:02:34.260 If the right antenna is still touching, it needs to continue turning. 00:02:34.260 --> 00:02:39.520 If the left antenna touches, it's run into a corner and needs to reorient itself so the 00:02:39.520 --> 00:02:45.799 new wall is on its right, the situation we dealt with the rotate-counterclockwise state. 00:02:45.799 --> 00:02:50.420 Finally, if both antennae are free, the ant should be in the state of the previous slide: 00:02:50.420 --> 00:02:55.900 standing parallel to the wall, so we return the WALL1 state. 00:02:55.900 --> 00:03:01.219 Our expectation is that the FSM will alternate between the WALL1 and WALL2 states as the 00:03:01.219 --> 00:03:02.640 ant moves along the wall. 00:03:02.640 --> 00:03:08.340 If it reaches an inside corner, it rotates to put the new wall on its right and keeps 00:03:08.340 --> 00:03:09.340 going. 00:03:09.340 --> 00:03:13.079 What happens when it reaches an outside corner? 00:03:13.079 --> 00:03:19.329 When the ant is in the WALL1 state, it moves forward and turns right, then checks its right 00:03:19.329 --> 00:03:23.480 antenna, expecting the find the wall its traveling along. 00:03:23.480 --> 00:03:25.961 But if its an outside corner, there's no wall to touch! 00:03:25.961 --> 00:03:31.319 The correct strategy in this case is to keep turning right and stepping forward until the 00:03:31.319 --> 00:03:34.849 right antenna touches the wall that's around the corner. 00:03:34.849 --> 00:03:39.500 The CORNER state implements this strategy, transitioning to the WALL2 state when the 00:03:39.500 --> 00:03:41.590 ant reaches the wall again. 00:03:41.590 --> 00:03:44.230 Hey, this might even work! 00:03:44.230 --> 00:03:46.250 Let's try it out… 00:03:46.250 --> 00:03:48.959 Meet the Roboant simulator. 00:03:48.959 --> 00:03:54.609 On the left we see a text representation of the transition table for the FSM brain. 00:03:54.609 --> 00:04:00.689 Each action line specifies an input pattern, which, if it matches, will set the next state 00:04:00.689 --> 00:04:03.980 and output signals as specified. 00:04:03.980 --> 00:04:08.499 This particular version of Roboant allows the ant to drop or pickup breadcrumbs, and 00:04:08.499 --> 00:04:13.609 to sense breadcrumbs it comes across - these inputs and outputs aren't needed for this 00:04:13.609 --> 00:04:15.310 demo. 00:04:15.310 --> 00:04:20.500 The input pattern specifies a value for the current state and antenna inputs. 00:04:20.500 --> 00:04:26.090 The simulator highlights the row in the table that matches the current inputs. 00:04:26.090 --> 00:04:33.409 As you can see, initially the ant is the LOST state with neither antennae touching. 00:04:33.409 --> 00:04:38.730 On the right is a map showing our green ant standing in a maze with blue walls. 00:04:38.730 --> 00:04:41.730 We can select several different mazes to try. 00:04:41.730 --> 00:04:46.630 To see the ant in action, let's click the STEP button several times. 00:04:46.630 --> 00:04:52.090 After a few steps, the ant hits the wall, then rotates counterclockwise to free its 00:04:52.090 --> 00:04:53.640 antenna. 00:04:53.640 --> 00:04:57.510 Now it starts following the wall until it reaches a corner, at which point it keeps 00:04:57.510 --> 00:05:02.530 turning right and stepping until it's once again in contact with the wall. 00:05:02.530 --> 00:05:08.660 Now we'll let it run and watch as the FSM patiently pursues the programmed strategy, 00:05:08.660 --> 00:05:13.500 responding to inputs and generating the appropriate output responses. 00:05:13.500 --> 00:05:19.040 With more sensors and actuators, you can see that fairly sophisticated behaviors and responses 00:05:19.040 --> 00:05:21.220 would be possible. 00:05:21.220 --> 00:05:27.790 In essence this is exactly what modern robots do - they too have FSM brains full of pre-programmed 00:05:27.790 --> 00:05:31.470 behaviors that let them perform their assigned tasks. 00:05:31.470 --> 00:05:32.300 Neat!