A Discussion About SLAM in Dynamic Environments [Wolf and

Download Report

Transcript A Discussion About SLAM in Dynamic Environments [Wolf and

A Discussion About
SLAM in Dynamic Environments
Wolf and Sukhatme, ICRA 2004
Aisha Walcott
Research Group Meeting
Feb. 20, 2009
Problem
• Real-world environments consists of objects of type:
Static
Low dynamic/Semi-static
High dynamic/Dynamic
Classical Map Representations
[Thrun, Burgad, and Fox, 2005]
Occupancy Grids
http://www.asl.ethz.ch/people/rtriebel/personal/classes/Mapping.pdf
[Moravec, 1988]
The Paper
Contributions
• Algorithm capable of differentiating static
and dynamic parts of the environment
• Map is union of a static occ. grid, S, and a
dynamic occ. grid, D: S U D
• Landmark-based localization using the
static map to distinguish moving landmarks
Problem Formulation
•Estimate the static and the dynamic maps, given all the
observations o, and the positions of the robot u,
•Robot position not used to update occupancy probability
•To differentiate static and dynamic information, incorporate previous
static map in St-1 estimation,
Current static map All observations
Current dynamic map
Previous static map
•Occupancy of S and D maps are mutually exclusive, rewrite as
two equations,
Static Map Update
• Inverse observation model considers both:
Observation, ot
Previous occupancy probability, St-1
Dynamic Map Update
• Maintain information about dynamic objects only
at the present time
Example: Map Update
Landmark-based Localization
• Problem- moving objects detected as landmarks cause
errors in localization
• Use static map to differentiate static landmarks from
dynamic
• Landmark is static if the occupation probability of that
region in the Static map is classified as occupied
Experimental Setup
• Simulation: Player/Stage
– Pioneer robots, SICK laser
– Robots as moving obstacles
– Obstacles moved out of field of view
– Obstacles had corner (landmarks)
• Real-world
– Hallways of USC
– Moved objects
– Three people walking around
Results: Simulation
SLAM Robot
Dynamic Entities
Landmarks
Results: Real Environment
Related Work
• Occupancy grids with different time scales
• Temporal occupancy grids
• Static-dynamic grid (includes Wolf and
Sukhatme)
N. Mitsou and C. Tzafestas, An Introduction to the Problem of Mapping Dynamic Environments,
Book Chapter on "Tools in Artificial Intelligence", I-Tech Education and Publishing, Vienna, 2008.