Nattee Niparnan ROBOT CONTROL Behavior Based Robotic Towards

Nattee Niparnan ROBOT CONTROL

Behavior Based Robotic

Towards Autonomous Robot �A robot that can “think” how to perform the task

Autonomous? � Able to do things by itself. � Robot Control System �A system that decide what / when / how to do a particular thing to achieve the given task

Hierarchy of Control � Reductionism Follow the white rabbit Get dress walk to the pub talk choose a shirt wear a shirt Move a hand to wardrobe

Robot = ? ? ? �“ A device that connects sensing to actuation in an intelligent way” Intelligent

Model-Based approach � Sense Plan Act

Model-Based approach � Understand the world � Planning according to the state of the world � Result in rules for actions �If … then … �. �.

Remember the Shakey?

Robot Control Issue � Model of the world? � Robust?

Problem of model based � It seems reasonable � Does it work well in practice? �Model can hardly be realized �Model based is more appropriated with structured environment �Parallel nature? �GIGO issue

Problem of model based � Example, �Self Charging ○ Walk to beacon ○ Engage charger approach maneuver ○ Plug-in ○ stop �What if we are near the charger?

Problem of model based � What if we are near the charger? �Does our plan cover this case? � Coupling between requirement �Usually bug prone � Model based is sometime “computer oriented”

Computer vs. Robot � All computers are equivalent (turing machine) � Any two robots are different

Truth about Robots have sensors that measure the aspect of external worlds � Robots have actuators that can act on the robot and on the world � The output of a robot’s sensors always includes noise and other errors � The commands given to a mobile robot’s actuators are never executed faithfully. �

Sensing For us (human)… � For them (robot)… �

Actuation � Electrical signal Physical quantity �Always has some error

Intelligence � Robot design + Robot’s Program + Robot’s environment = Robot’s Intelligence

Mobile vs. Immobile Robots

Mobile vs. Immobile Robots Mobile Immobile Unknown world Highly structured world Dynamic Environment Static Environment Localization and mapping problem

Example � Collecting a puck and put it into light

Tasks � Show gizmo and collection tasks in Bsim � What we have as a low level command?

Behavior based control � What are used in Gizmo

Example of Behavior Based

Behavior based robotics

Behavior based robotics � Reflexive �Shortest time from sense act � Carefully engineered the reflex to actually perform the task

Principle � World = what robot sees � Plan less �Check Act more � Be highly adaptable to change �Agility?

Intro to Control

Lower Level Control � Given desired output � Find input that yield such output

System Input U Black Box (grey box) System Output Y

Control � We hardly understand our system � The mathematical model “approximately” describe the system � There always be some error � There might be some unknown rule!

Example � Do we know the speed of motor �If we apply some specific voltage? �Without actually measuring? �i. e. , forward computation � We have all theory, right?

So what? � If we don’t really understand the system �How do we calculate U for given Y? �I want my motor to spin at 200 rpm �What voltage should I put? �Who knows?

The Solution � Control System �Open loop �Closed loop

Control System � Open loop

Open Loop � Just supply input �From the model � Example �Light bulb �Electric fan

Open Loop � Neglect input �Hence, does not adapt itself to the world �Very simple �Easily failed � Work perfectly if we know perfect model of the system �Which is not usually the case

Control System � Open loop

Control System � Closed loop

Feedback Control � Very important to accommodate error � We already did that all the time �Your body �Your brain �Your eco system

Trichotomy Measurement � Yes � More � Less

Proportional Controller � Feedback with degree � Include error of the output �Multiply by the proportion of the error ○ i. e. , gain of the control

Closed-Loop Control Example � Position Control

BSim � Gizmo task

Problems � Slow to adapt � Solve by increase gain

BSim again � Try to increase gain

Control System Catastrophe

Latency Problem � Result from the control does not actually reflect the current state � Lead to instability �Sometime to catastrophe

Control System Stability

PID Controller

PID Controller � Proportional Part �Normal close loop � Differential Part �Adjust input by the differential of the error � Integral Part �Adjust input by the

Tuning PID � Adjust P to converge

Tuning PID � Add D to solve overshoot

Tuning PID � Add I to solve Steady State

Tuning PID � Actually an black-art �Tuning the knob has highly coupling effect �Let’s try it

Tuning PID summary Change in Rise Time Overshoot S-S Error parameter Settling Time Increase P Less More Less Minor Increase D Less More Eliminate More Increase I Less Minor

Saturation, Backlash, Dead Zone

Saturation, Backlash, Dead Zone

Open Loop Enhancement � Parameters � States

Bang-Bang Controller

Hysteresis

More control scheme � Feed forward � Predictive � Adaptive

Dynamic System � Even if we perfectly understand the system, it is still not trivial to achieve good control

Example � We can solve for u for a given y Input u System with perfect knowledge Output y

Example � Taken from Stephen Boyd class � Input 2 dimension � Output 2 dimension � x˙ = Ax + Bu, y = Cx, x(0) = 0 �Differential equation � Says, we want y = (1, -2) � We can solve u to be (-0. 63, 36)

Use the simple

Example

Final Words � You cannot learn how to program robot from looking at this slide � BSim? �What works well in sim does not always works well in practice � Let’s do LEGO!

Introduce Lego Mindstorm

Example � Show example of Roverbot �Pushbot �Guardbot �Explorer �Mozart

Assignment � Pick a robot from LEGO kit � Do something with it � It’s 10%