disputedip.com

Well it made its first corner, video:

Go to http://www.youtube.com/watch?v=5a4zVuxzEdg for the HD version.

And here is the data from that run:

The todo list is as follows:

• Retune controller (there is some nasty overshoot and the control signal itself is oscillating)
• Calibrate the speeds - left and right wheel have different speeds (:() so need to go an work out the difference
• Automate this, cause it took all day to get to this point

Please note that this is a ‘jigged’ callibration - the long straight allows me to guestimate the required speed differences and ticks per meter, so not too precise at this point in time.

As its exam time I have been carefully avoiding revision with a mix of counterstrike and micromouse work. New PCBs are 95% ready to be ordered and so my focus has turned to the controllers.

After implementing a fixed point maths system for the controller (10.5bits signed for those who care) and ziegler-nichols tuning the PI controller I had a one wheel mouse - putting a bit more effort in I got it to two wheels - yay.

Tuning it was a bit of a chore - so automation is king and I was able to produce this graph (yes, no axis labels, minus 2 marks for me, x is the time, y (going back into the scene) is as the gain increases and z (going up) is the velocity for said gain:

The idea is to find the point where oscillations occur, the above plot was far too noisy so a 16 point moving average was applied:

I got a value of about 1.2, but it is questionable, will compare against some simulated controllers to see if it seems a likely value - it does however work well enough to control the motors!

Now the results:

It is nice having onboard bluetooth for this sort of debugging, anyhow, things look nice - the integral term does push things out of whack when going from zero or getting down to zero - but possibly an artefact of the controller being ill tuned - there is still a large overshoot to a step input which is not what you would expect from a z-n tuned controller…

Video, go to http://www.youtube.com/watch?v=AYmkJvthVJw and click on view in HD at the bottom! or watch it in small size bellow:

Next time it will be slower and have a straight line controller built in, after that its onto cornering!

Very bad video quality - but the mouse moves:

Will get a better camera in on Monday, but yay it moves. This is a very basic movement - accelerate to ~20% maximum speed, stay there and start to decelerate after 1s.

It moved very roughly, 0.5m in 1sec, making its theoretical (ish) top speed about 2.5m/s (when loaded). Although that metric does not take into account anything, so could be higher or lower :p.

Buttons are cool, so are menu’s. So with the help of a button, a few leds and a dspic I have a menu system (also note that this is blocking, the idea is that its used to start / stop the robot, so the robot does not move while the menu is active), first lets look at the switch circuit and then at the dspic set up (c30 compiler, using a dspic33fj128gp804):

Schematic for a switch for a dspic

/* Setup the IO Pin Assignment */
__builtin_write_OSCCONL(OSCCON & ~(1<<6));

	RPINR7bits.IC1R = 5;	// Switch Input Capture

__builtin_write_OSCCONL(OSCCON | (1<<6));

/* Switch Input Capture Setup */
IC1CONbits.ICM = 0x02;	// Capture on falling edge
_IC1IF = 0;
_IC1IE = 1;

/* CPU Time Clock */
T1CONbits.TCKPS = 0x02;	// 40.5504Mhz / 64
TMR1 = 0x00;
PR1 = 634*4;			// Int at 1.000631313ms*4
_T1IF = 0;
_T1IE = 1;
T1CONbits.TON = 1;

The code is fairly simple: first attach the switch input to the input capture (ic1) peripheral, set up ic1 to trigger an interrupt on the falling edge and then there is the set up for my main timer loop, this occurs at ~250Hz, and yes I am using a weird value crystal.

Next lets look at the global variables (well global to the scope of the interrupts)

// Switch input vars
int sw_counter = 0;
int sw_on = 0;

// Menu vars
int menu_counter = 0;
int menu_on = 0;
int menu_value = 0;

Next the interrupt for the input capture routine

void __attribute__((__interrupt__)) _NOPSV _IC1Interrupt(void)
{
	_IC1IF = 0; 		// Clear the interrupt
	_IC1IE = 0;			// Disable the interrupt
	_RED = 1;   		// Show the user the button pressed
	sw_on = 1; 		// Enable the debounce counter

	// If the menu is on already, increment the menu value
	// Otherwise turn the menu on and set the menu value to
	//default
	if (menu_on)
	{
		menu_val ++;
		menu_counter = 0;
	}
	else
	{
		menu_on = 1;
		menu_val = 0;
	}
}

And then the interrupt for the timer:

void __attribute__((__interrupt__)) _NOPSV _T1Interrupt(void)
{
	// Handle switch
	if (sw_on)				// If a switch 'event' is ongoing
	{
		sw_counter ++;
		if (sw_counter > 50)  	// 50x 1/250 = 200ms debounce
		{
			sw_on = 0;		// Switch no longer on
			sw_counter = 0;	// Counter to zero
			_IC1IF = 0;		// Clear the bounces
			_IC1IE = 1;		// Enable more switches
			_RED = 0;		// Led off
		}
	}

	// Menu
	if (menu_on)			// If menu is on
	{
		menu_counter ++;	// Increment counter

		// Leave menu active for 1s after last click
		if (menu_counter > 250)
		{
			menu_on = 0;	// Menu off
			menu_counter = 0; // Reset

			// Menu complete - do stuff
			switch (menu_val)
			{
				case 0:
					_RED = 0;
					Nop();	// Required if RED and GREEN
							// On same port
					_GREEN = 1;
					break;
				case 1:
					_RED = 1;
					Nop();	// Required if RED and GREEN
							// On same port
					_GREEN = 0;
					break;
				default:
					break;
			}
		}
	}
}

And thats it, google _NOPSV to find the macro for that (sometimes required to compile using c30). Also _RED and _GREEN are port bits, so something like #define _RED _RB1 to make _RED = 1 turn Port B pin 1 on. All quite useful, maybe…

Today was a very unproductive one. Slept in, managed to do nothing till 2, then spend hours trying to work out why something had broken on my mouse - it was me, it was running 4 times slower because I put the line of code to tell it to run 4 times faster in the wrong place, that followed by a c# bottleneck - c#.net cannot handle 1mbit rs232 too well, its running fine at about 20% capacity, but between there and 50% capacity something goes very wrong. That lead to hours of no progress. And finally I though I may have broken one of my motors, the expensive ones, but it was fine, a bad connector, or maybe a break in the cable (think it was the former as after a lot of wiggling I cant detect a break). It must work - look at the graph bellow result.

A very messy desk

Graph if step input to motor

The ‘rough’ steady state speed of this input is about 2m/s, mwhahaha, lets hope there is some torque at 0.3m/s otherwise testing will be interesting….