Group items matching

in title, tags, annotations or url

how to interpret sensor data into a mathematical form readable by computers

There are only 3 steps you need to follow: Gather Sensor Data (data logging) Graph Sensor Data Generate Line Equation

Some sensors (such as sonar and Sharp IR) do not work properly at very close range

robot learning

variable delay

short-term memory

Droidmakr (Cliff Boerema) came up with an interesting idea for a light-tracking head with a form of peripheral vision. As often happens, the circuit turned into something different -- a photopopper:

All done with a single 74HC14 (the '240 being a motor driver).

I tried the same setup with the 74*240 (with an extra inverter per motor) and 7404, but the 74HC14 seems to work best.

"Exposing a Flaw: Shoot-Through" describes the serious problem with that circuit, especially when pulsed

Above is an improved version of the circuit, which is now PWM compatible. PWM, coast mode, and the capability to avoid shoot-through are provided by adding a fifth MOSFET (labeled Q5) to the source/ground connections of Q1 and Q3.

Pleo Technical Specs: Ugobe LifeOS 32 bit Atmel ARM 7 microprocessor - The main processor for Pleo 16 bit sub processor - The processor dedicated to the camera system (4) 8 bit processors that provide the low-level motor control for the servos (35) Sensors including a camera custom designed to fit into Pleo’s very compact body. (4) foot-switches to detect footfalls and being picked up - assists with spatial orientation. (12) capacitive touch sensors (4) legs, (4) feet, back, shoulder, head, chin (2) microphones for directional sound detection (14) “Force” sensors, one per servo, to recognize abuse through force feedback joints. Orientation/tilt sensor IR transceiver for bidirectional data communication with other Pleos. IR interrupter for detection of objects in Pleo’s mouth (14) motors. Standard low voltage DC motors (150) gears and clutches Rechargeable NiMH battery pack USB port with mini USB connector SD/MMC memory card slot

CALEB CHUNG: Of course we could have used micro-servo motors to accomplish the motion of Pleo, but we aren’t able to use expensive motors. So we had to engineer it with a high-speed motor with high gearing and no backlash for control purposes and have it all fit within the muscle envelope of Pleo.

So what we did was go after a lot of ethology research. How do animals really handle the complexity of their environment? We built a virtual brain—a whole system that decides how Pleo will react in various situations.

CALEB CHUNG: Pleo will reset thresholds and adjust his idea of what he thinks is normal. Let’s say you get Pleo and you take him home to your shag carpet. When Pleo walks, the carpet will drag on his feet. So his force feedback sensors will realize that he is spending too much energy to walk around. Pleo will try different things to reduce the energy spent. Eventually, he will have the idea to step higher. Your Pleo compared to my Pleo will walk with a higher step.

The algorithm used here is a simple Q-learning algorithm

motors commercially available do not normally have a desirable speed to torque ratio (the main exception being servos and high torque motors with built in gearboxes)

With gears, you will exchange the high velocity with a better torque.

the larger gear will move more slowly than the smaller gear, but it will move with more torque.