Demonstrations |
2012-04-30 |
SBL Visual Intelligence |
Application of SBL to verb recognition in video data streams. The robot is the observer
and the actions are carried out by humans.
|
|
Author(s) |
NHK |
|
2011-08-08 |
SBL on Gatchan NHK |
Demo/Intro of SBL on SuperBot for the Japanese NHK Gatchan show. |
|
Author(s) |
NHK |
|
2011-03-28 |
Adapting to unpredicted changes |
The robot explores the office through random executions of the left and right actions. A
goal is reached by planning with the learned model. A simultaneous unpredicted sensor, action,
environment and goal change is introduced by rotating the camera, toggling the left and right
actions and moving a goal book. SBL detects surprises, repairs its model and tracks the goal.
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2010-09-01 |
Learning to Turn Left/Right in an Office |
SBL learns the result of the turn left and turn right actions in the office. It it
tasked to match the goal scene using only these actions after learning them.
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2010-08-24 |
Learning Forward/Backward in an Office |
SBL learns the result of the forward and backward actions in the office. It it tasked to
match the goal scene using only these actions after learning them.
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2010-08-22 |
Real-World Office Environment |
The real-world office environment used for testing autonomous navigation. This consists
of a SuperBot module being controlled by a laptop via Bluetooth in a typical office room with a few
magazines/books placed scattered around. Objects are detected using SURF features.
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2009-11-15 |
Unpredicted sensor change |
Demonstration of learning before and after an unpredicted camera sensor change. |
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-08-29 |
Fault Tolerance Experiment 5 |
The improved Surprise-Based algorithm learning with autonomous sensor & actuator
coupling, rule forgetting and feature relevance. The left turn and right turn actions are toggled
after a model has been learned so as to simulate actuator failure.
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-08-27 |
Fault Tolerance Experiment 4 |
The improved Surprise-Based algorithm learning with autonomous sensor & actuator
coupling, rule forgetting and feature relevance. The camera is flipped by 180 degrees to simulate
sensor failure.
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-08-26 |
Fault Tolerance Experiment 3 |
The improved Surprise-Based algorithm learning with autonomous sensor & actuator
coupling, rule forgetting and feature relevance. A random valued sensor with no correlation to the
environment is added to test feature relevance
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-08-25 |
Fault Tolerance Experiment 2 |
The improved Surprise-Based algorithm learning with autonomous sensor & actuator
coupling, rule forgetting and feature relevance. A constant valued sensor is added to test feature
relevance
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-06-07 |
SBL experimental setup |
A flyby of the surprise-based learning experimental environment and robot |
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-06-07 |
SBL sensor toggle experiment - external |
An external camera view of the an SBL experiment where the camera is flipped 180 degrees
after learning for a while. Targetting after learning for a short while is shown, followed by
targetting after sensor toggle and relearning
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-06-07 |
SBL sensor toggle experiment - data |
An internal data view of an SBL experiment where the camera is flipped 180 degrees after
learning for a while. Targetting after learning for a short while is shown, followed by targetting
after sensor toggle and relearning
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-06-06 |
SBL motion prior to learning - line |
The target is the red wall and white floor seen from a particular distance. The robot
has not learnt how to move backwards to accomplish this.
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-06-06 |
SBL motion prior to learning - turn |
The target is the green wall. The robot has not learnt how to turn to accomplish this.
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-06-06 |
SBL motion prior to learning - corner |
The target is the corner of the red and green walls. The robot has not learnt how to
turn to accomplish this.
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-06-06 |
SBL motion after learning - line |
The target is the red wall and white floor seen from a particular distance. The robot
has learnt how to move backwards to accomplish this.
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-06-06 |
SBL motion after learning - turn |
The target is the red wall. The robot has learnt how to turn to accomplish this. |
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-06-06 |
SBL motion after learning - corner |
The target is the corner of the red and yellow walls. The robot has learnt how to turn
to accomplish this.
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-06-06 |
SBL motion with flipped camera - line |
The camera is flipped while the robot was learning. The robot is unable to move to
target due to surprises
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-06-06 |
SBL motion with flipped camera - turn |
The camera is flipped while the robot was learning. The robot is unable to turn towards
the green wall without further learning
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-06-06 |
SBL motion with flipped camera - corner |
The camera is flipped while the robot was learning. The robot is unable to turn towards
the corner of the blue and red walls
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-06-06 |
SBL motion with flipped camera and relearning -
line
|
The camera is flipped while the robot was learning and is allowed to learn for a while
longer. The robot is able to learn move backwards to see the green wall and white floor at a
distance.
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-06-06 |
SBL motion with flipped camera and relearning -
turn
|
The camera is flipped while the robot was learning and is allowed to learn for a while
longer. The robot adapts to turn towards the blue wall.
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-06-06 |
SBL motion with flipped camera and relearning -
corner
|
The camera is flipped while the robot was learning and is allowed to learn for a while
longer. The robot adapts to turn towards the corner of the green and yellow walls.
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-05-12 |
SBL Complementary Rules & Multi Stage Planner |
The complementary rules have been padded to accomodate ABSENT explicitly. Also the
planner has been modified to find a route to the targets in the first stage, move to remove
fiducials that are not a part of the target scene and finally plan to adjust the sizes
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-04-10 |
SBL sensor toggle |
The camera image has been flipped along the vertical axis. SBL is still able to learn
and track a target.
|
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-04-09 |
SBL tracking target |
Learning a world model and using it to track a target which is set during runtime. |
|
Author(s) |
Nadeesha Ranasinghe |
|
2008-04-07 |
SBL world model |
This video shows how SBL creates a compact world model with very few surprises. |
|
Author(s) |
Nadeesha Ranasinghe |
|