AIMA 25.1-25.4

Introduction to RoboticsPresented by Derek Colla[additions by Simon Levy]

25.1: Introduction

Robot – active, artificial agent whose environment is the physical world

Autonomous robots- make decisions of their ownWe will focus on these

Five properties of environments

Inaccessiblesensors are imperfect, and can only

perceive stimuli close by

Nondeterministicrobot needs to deal with uncertainly,

because problems arise (broken parts, batteries run low, etc.)

Five properties of environments

Nonepisodic the effects of an action change over time,

so robot must handle sequential decision problems and learn.

Dynamicrobot must know when it is worth

deliberating, and when to act immediately

Five properties of environments

Continuous In the real world, states and action are

drawn from a continuum of physical configurations and motions. This makes it impossible to enumerate the set of possible actions.

25.2: Tasks: What are robots good for?

Manufacturing and materials handling traditional domains. Robots used in

manufacturing are not usually autonomous.

Gofer robotsmobile robots (mobots) that serve as

couriers and security guards in hospitals and office buildings

currently becoming widely used

25.2: Tasks: What are robots good for?

Hazardous environmentsClean-up and maintenance, rescue

operations, and space and deep-sea exploration (Mars Rover)

Telepresence and Virtual RealityThings such as having a glove with a

remote sense of touch

Sojourner

25.2: Tasks: What are robots good for?

Augmentation of human abilities

25.3: Parts: What Are Robots Made of?

Linkssimilar to the forearm or upper arm

Jointssimilar to the elbow or shoulder

Effectors

Any device that affects the environment, under the control of the robot

Actuators converts software commands into physical

motion Ex: electric motor

Degrees of freedom describes how many independent ways the robot can move (up/down; left/right; forward/back)

Effectors

Effectors used in two main ways: Locomotion

Changing the position of the robot within its environment

Manipulation Moving other objects in the environment

Locomotion

Types of Legged Locomotion Statically stable

A robot that can pause at any stage during its gait without tumbling over

Slow and energy inefficient Dynamically stable

A robot that would crash if forced to pause, but does well as long as it keeps moving

Usually use a hopping motion

Static Stability

Ocotopod, by Prof. David Livingston (VMI)

Dynamic Stability

Hopper, by Prof. Marc Raibert MIT Leg Lab (http://www.ai.mit.edu/projects/leglab/)

Dynamic Stability

Hopper, by Prof. Marc Raibert MIT Leg Lab (http://www.ai.mit.edu/projects/leglab/)

Locomotion

Wheel or tread locomotion is the most practical for most environments More efficientEasier to buildEasier to program

Thought question: So why no wheels in nature???

LocomotionImportant distinction between how actuators move, and what effect these motions do to the environment Car Example

Turn wheel---Change direction of the car

Locomotion

If the controllable degrees of freedom is less than the total degree of freedom, than the robot is nonholonomic

The larger the gap, the harder it is to control the robot

If the controllable degrees of freedom are equal to the total, then the robot is holonomic These have a high mechanical complexity

Manipulation

Kinematics The study of the correspondence between the

actuator motions in a mechanism, and the resulting motion of its various parts (same word as “cinema” : one frame at a time)

Rotary motion Rotation around a fixed hub

Prismatic motion Linear movement, as with a piston inside a

cylinder (prism = general elongated polygon, e.g., triangular glass prism to refract light)

Manipulation

Manipulation

Six degrees of freedom A free body in space has six degrees of

freedom (three for x-y-z position, three for orientation), so six is the minimum number of joints a robot requires in order to be able to get the last link into an arbitrary position and orientation

End effector at the end of a manipulator Can be screwdriver, suction cup, etc.

Sensors: Tools for perception

Proprioception Sense to tell a robot where its joints are

Humans have this as well

Encoders fitted to the joints provide very accurate data about joint angle or extension Allows robots to have a great degree of

positioning accuracy, much better than humans

Sensors: Tools for perception

Repeatability How your positioning improves given more

than one try

Odometry measures length of movement, can be

error prone due to slippage

Sensors

Force sensors Needed for things like scraping paint off a

window

Compliant motions Moving along a surface while maintaining

contact with a fixed applied pressure

Sensors

Tactile sensing robotic version of the human sense of

touch

Sonar Sound navigation and rangingUsed mostly for fast collision avoidanceAlso can position obstacles

Difficulties with mapping though

Sensors

Camera dataDomain constraints can help simplify things

for special-purpose robotsStructured light sensors

Project their own light source onto objects to simplify the problem of shape determination

25.4: Architectures

PurposeDefines how the job of generating actions

from percepts is organized

Classical ArchitectureHierarchy

Intermediate-level actions and low-level actions incorporated to reduce errors

Compiled results into macro-operators to allow the robot to learn.

Architectures

Situated automata The principle drawback of the classical

view is that explicit reasoning about the effects of low-level actions is too expensive to generate real-time behavior.

A situated automaton is a machine whose inputs are provided by sensors connected to the environment, and whose outputs are connected to effectors. The s.a. has only a limited (finite) number of possible states.

Situated Automata

Very efficient implementation of reflex agents with state.

Implements specific laws, such as the laws of physics given to the compiler, and uses those to make propositions

Behavior-based Robotics

Based on the idea that the agent design can be decomposed, not into functional components such as perception, learning, and planning, but into behaviors such as obstacle avoidance, wall-following, and explorations.

Contrasts with planning approach : robot thinks for a long time about how to accomplish a goal, then acts.

Behavior-based Robotics

Genghis, by Prof. Rodney BrooksMIT AI Lab / iRobot Corp (Roomba)

Behavior-based Robotics

Genghis, by Prof. Rodney BrooksMIT AI Lab / iRobot Corp (Roomba)