Logo-white-medium.jpg

asr_ros_pkgs is a software framework which impements the approach 'Active Scene Recognition (ASR)' and has been developed at the Chair of Prof. Dr.-Ing. R. Dillmann, Humanoids and Intelligence Systems Lab (HIS), Karlsruhe Institute of Techology (KIT), Germany. ASR is an approach for mobile robots to recognize scenes in distributed and cluttered indoor environments. It combines scene recognition and object search: Information about partially recognized scenes is used to guide the subsequent search for missing objects. In implementing ASR, asr_ros_pkgs provides both

  • A scene classifier: A hierarchical data structure we introduce as 'Implicit Shape Model trees' together with learning and recognition algorithms.
  • An integrated software architecture: A hierarchical state machine for ASR, including components for both scene recognition and object search.

The scene classifier can be used independently of ASR for what we call Passive Scene Recognition (PSR). As PSR, we define scene recognition with an immobile observer.

This project is of use for any researcher who intends to estimate the state of a robot's environment based on both the objects present in the environment and the 6-D spatial relations between the objects. Such estimates can be used for robotic decision-making, e.g., in the context of action planning.

As a starting point for

The research project 'Learning the Context in Programming by Demonstration of Manipulation Tasks', during the course of which asr_ros_pkgs was created, has been a 3-year research grant funded by the German Research Foundation (DFG).

installation.png

  • The install instructions will guide you through the installation of ASR on Ubuntu 16.04, click here.

all_components.png

active_scene_recognition.png

  • The active_scene_recognition consists of different packages, on a high level of abstraction, to handle the 3D-object-search. The main goal is to recognize which scene(s) are present in the current search space. The scenes consist of objects and the relative relations between them.
  • All active scene recognition packages

control.png

  • The control package provides ROS access to actuators that both of our experimental setups (dome and robot) have in common. In particular, we provide a general-purpose ROS driver for FLIR pan-tilt-units.
  • All control packages

dome.png

  • Packages which are related to the kinematic chain of the experimental setup „dome“ at our lab. This setup consists of a cage of aluminum profiles on which two actuated robot heads are mounted as well as of a magnet-tracking system "Flock of Birds“ and a pair of Cybergloves.
  • All dome packages

asr_msgs.png

  • This package contains all messages that are particular to our Active Scene Recognition system. These messages make up the interfaces between the different collaborating components of our software architecture.
  • asr_msgs package

robot.png

  • The robot package provides ROS access to navigation-related actuators and sensors, the kinematic chain and a gazebo simulation of our mobile robot (resp. our experimental environment „mobile robot). Moreover,, it provides a general-purpose local planner, following the „follow-the-carrot“ paradigm.
  • All robot packages

perception.png

  • The perception package wraps up low-level and high-level perception packages from RGB-D, laser and stereo-camera interfacing, over 6-DoF object localization to (passive) scene recognition. Passive scene recognition is performed with Implicit Shape Model trees, while 6-DoF object localization can be done by the help of fiducial markers or on the basis of the HALCON library.
  • All perception packages

utility.png

  • The utility package wraps up miscellaneous support functionalities, required by a variety of our packages.
  • All utility packages

Literature.png

Wiki: asr (last edited 2019-11-11 17:42:46 by FelixMarek)