1/84
Looks like no tags are added yet.
Name | Mastery | Learn | Test | Matching | Spaced |
---|
No study sessions yet.
situation
adapting and reacting differently to the changing environment (may include learning)
acting
the ability to bring changes in the world
being useful
the ability to achieve some goals
"learning by building"
the idea that one can truly understand something only by building, creating it
emergent behaviour
complex, often unexpected patterns, actions, and behavior that arise from relatively simple rules and/or architecture
the Tortoises
made by WGW, first electromechanical robots. Emergent behavior: 2 sensors, 2 actuators, programmed reactive control
robot
an autonomous system embodied in the physical world, is able to interact, and sense input from it to achieve goals
autonomous
acting based on its own decisions, not being controlled by humans
teleoperated
controlled externally by humans
embodiment
needs to obey laws of physics, and needs to be capable of maintaining itself in the physical world
Braitenberg Vehicles
14 constructed vehicles with more and more "brainpower". Idea: understanding intelligence by creating it.
Vehicle 1
getting around, likes cold, avoids heat; 1 sensor, 1 motor. In nature: Euglena
Vehicle 2
detecting light, each light sensor is connected to a motor on the same side or the opposite side of the vehicle. It uses 2 sensors and 2 motors.
Vehicle 2 Fear
each sensor is connected directly to the motor on the same side, the vehicle appears "fearful" toward the light
Vehicle 2 Aggression
each sensor is connected to the motor on the opposite side, the vehicle behaves in a way that appears to be "aggressive" towards the light
Vehicle 3
the sensors are connected to the motors with inhibitory wiring, so instead of speeding up the motor in response to stronger sensor input, stronger light input causes the motor to slow down. It also uses 2 sensors and 2 motors
Vehicle 3 Love
each sensor is connected to the motor on the same side with an inhibitory connection, the vehicle appears to "love" the light
Vehicle 3 Explorer
each sensor inhibits the motor on the opposite side, the vehicle moves away from strong light sources but may approach and linger in dimly lit areas
Vehicle 4
combines multiple sensory inputs, each linked to different types of stimuli, resulting in more complex behaviors
Vehicle 1-3
Limited autonomy, purely reactive, yet complex behaviors arise from simple excitatory or inhibitory wiring; shows attraction, avoidance, and exploration.
Vehicle 4-6
Increased adaptability and dynamic response; responds to multiple stimuli, adapts to frequent exposure, and begins storing recent data, showing habituation and anticipation.
Vehicle 7-8
Goal-oriented and spatially aware; follows gradients toward goals and learns paths, suggesting early goal-seeking and path memory.
frame of reference problem
concept in robotics, AI, and cognitive science. It addresses the challenge of understanding how different "observers" (robots, systems, or even people) perceive and interpret the same situation differently depending on their unique position, orientation, or context. 3 main points (in this context): - humans detect human-like motivation/emotion, hereby creating false connections - a behavior of a robot is explained not just by its internal mechanisms, but its interaction with the environment (outside world) - complex behavior doesn't necessarily mean complex (internal) mechanisms
Shakey (1966-72)
early AI-inspired robot, that works in carefully prepared lab conditions. It focuses on reasoning and planning
sensors
to be able to perceive the real world
state
description of a robot and its environment at a given point in time (eg. position, orientation, velocity)
state space
consists of all possible states a robot can be in
sensor space
aka perception space, all possible sensor readings a robot can have
states can be
Observable/Partially observable/Hidden, Continuous/Discrete
effectors & actuators
components of a robot that allow it to move (The effector is the arm, and the actuator is the muscle(s) moving it)
passive activation
only potential energy is used for moving (eg. the stepping robot)
active activation
uses external energy for moving (ex. NEO)
examples of actuators
electronic motors, hydraulics, pneumatics, reactive material
DC motors
the more power it gets, the faster it turns (Fe->Fm)
Servo motors
turn shaft to a specific position, usually made from DC motors
control
("brain") needed for an AS, combines sensory input with actuator output
DOF
Degrees Of Freedom - min nr. of coordinates needed to completely specify the motion of a mechanical system
Transitional DOF
x,y,z
Rotational DOF
roll, pitch, yaw
holonomic
controllable DOF
non-holonomic
controllable DOF < Total DOF (eg. car - 2DOF (x,z))
redundant
controllable DOF > Total DOF (human arm - 7 DOF)
robot locomotion
the way a robot moves from one place to another
static stability
stands still without falling (spider, tripod)
dynamic stability
needs to actively balance to not fall (pogo stick)
COG
center of gravity (a body is stable when the COG lies within the polygon of support)
gait
particular way a robot moves, characterized by the sequence of lifting and lowering legs and placing feet on the ground (eg. tripod, ripple) (possible gaits depend on the number of legs.)
robot perception
senses the environment (exteroception), and itself (proprioception)
active sensor
emits signal that interacts with its environment (sonar sensor)
passive sensor
measures physical properties without direct interaction (thermometer)
simple sensor
provides 1D input, that needs little to no processing (switch)
complex sensor
usually provides multidimensional input that requires sophisticated processing (camera)
sensor fusion
improving the usefulness of a sensor by combining it with other sensors
filtered sensing
applying knowledge about the world when processing a signal perceived by a sensor, resulting in simpler processing, and more accurate signal detection
Genetic algorithm (GA)
an approach to learning that is based on simulated evolution (unlike other ML algorithms)
GA processes
random solution, 2. evaluation, ranking 3. creating crossovers, offspring, 4. iterating this process until some goal is achieved
GA components
problem, fitness func., initial pool, evolution strategy (eg. MAXONE)
elitism
ensuring that the solution quality obtained will not decrease, and that the best solutions of a generation will carry over to the next
Genetic Operations
the operations that recombine and mutate selected members of generation Gx
crossovers
trying to simulate biological reproduction
mutation
adding randomness, noise, in hopes of creating a better solution when creating children
natural selection
probabilistic approach is used to decide which solution to keep (trying to resemble natural selection)
Genetic programming (GP)
extension of GAs for creating computer programs
controller
the robot's brain, that's what allows it to be autonomous
optimization
finding the best solution
control architecture
makes it possible to program a robot in such a way that it will produce the desired behavior
deliberate control
"Thinking hard" (eg. Mr. Vogt preparing for the lecture)
SPA architecture
3 steps of deliberate architect; Sense, Plan, Act
reactive control
"Don't think, react!" planning is removed to get fast results (eg. Braitenberg vehicles)
hybrid control
"Think and act separately, in parallel" combines deliberate and reactive control (eg. medicine delivery robot)
offline planning
prepare for situations that might come up (eg. preparing for an exam)
online planning
planning while task is performed (eg. writing the exam)
behavior-based control (BBC)
"Think the way you act!" biologically inspired (eg. humanoid robot)
subsumption architecture
hierarchical system, uses bottom-up design. The best-known architecture for BBC and reactive control
centralized representation
used by deliberative and hybrid control (global map), and centralized reasoning is used to solve the problem (planner to find path)
decentralized representation
it's distributed over behaviors' structure (reasoning is also distributed)
neural network controllers (NN controllers)
can be used by all types of control architectures
Robotic Operating System (ROS)
it consists of many systems, devices, and users that need to interact reliably
ways to organize computer systems
Distributed, parallel, hardware defines this choice
node
a single-purpose program that publishes or subscribes to information on a certain topic (eg. bumper message)
message
timestamped piece of information on a certain topic (eg. bool bumper_pressed)
Symbol grounding problem
symbols must be grounded in robot-environment interaction (eg. beer) (Newell & Simon '76 PSSH). It highlights the need for robots to have a robust mechanism to link abstract symbols to tangible objects and actions in the real world, enabling them to understand and execute commands accurately.
Referential uncertainty
everyone learning word-meaning mappings needs to deal with referential uncertainty (eg. Gavagai!)
effectors
components of a robot that allow it to interact with the physical world (eg. the arm)
actuator
the mechanisms that enable the effectors to execute an action or movement