Predict Human Intention in Visual Observations of Hand/Object Interactions Dan Song (responsible for the module: Intention Prediction)
Video of research paper published in ICRA 2013
First, a set of objects on the table are recognized with their poses estimated. Then the human hand pose and joint articulations are tracked with a Kinect sensor. The final stabilized grasp configuration is used as input to a Bayesian network task model to reason about the intention of the human: to hand-over, to pour, and/or to hammer.
Note that the task model is trained using a simulated human hand model, and here we show it can successfully infer tasks on real human grasps.
Alternatively, you may use the video download links: .mp4
Task-based Grasp Adaptation on Multiple Robot Platforms Dan Song (responsible for the module: Task-based Grasp Inference)
Video of research paper published in SYRACO 2012
This video shows a demo resulted from integration of multiple modules developed during the EU project GRASP.
Two robot platforms are: Armar Humanoid at KIT, and TOMBATOSSALS humanoid torsal at UJI.
Alternatively, you may use the video download links: .mov
2011
Task-based Object Selection and Grasp Planning in the OpenGRASP Simulator. Dan Song, Beatriz León, D. et. al.
The robot is the Armar Humanoid. The demo assumes the object pose and category are known. The robot selects the object and pre-planned grasps according to an assigned task by users.
Alternatively, you may use the video download links: .mov
2010
Data Generation Process for Learning Task Constraints of Robot Grasping using Graphical Models. Dan Song and Kai Huebner
Video showing data generation process supporting publications in IROS'10, RSS'10, ECCV'10, Humanoid'10, ICRA'11, IROS'11, ICRA'12, IROS'13 and ICRA'13.
The open-source software supporting data generation is: BADGr.
Alternatively, you may use the video download links: .ogg, .mp4, .avi
Published by: Dan Song <dsong@kth.se>
Updated 2013-06-06
