Demos

Demos

Demonstrating close Human-Robot collaboration in a shared workspace that is monitored by GPU-Voxels:

Here we demonstrate the Euclidean Distance Transformation feature of GPU-Voxels which allows you to derive 3D Distance Maps from pointclouds with high performance. In the video a flying drone plans a path through a environment, sensed only with on-board sensors. From the data a distance map is built on the fly with 30 Hz. So for every Voxel a simple wavefront planner can query the closest obstacle (plus its distance) and find a path that keeps a safety margin from the walls and only chooses openings that are large enough for the drone:

This two videos show our first tests on our HoLLiE Robot. On-board RGBD-Data is used to monitor the Swept-Volume of the planned trajectory. When collisions are detected, replanning is triggered. The process is fast enough to happen on the fly, without having to stop the robot.

 

Here we show the combination of RGB-D Scene Flow analysis and GPU-Voxels to realize predictive collision detection.

In this video we demonstrate the population of a GPU Octree with the Pointcloud data of two depth-cameras.

The following videos show the results of two planners that exploit the capabilities of our Voxel-Collision-Checking engine. Instead of evaluating single robot poses during planning, Swept-Volumes of whole motion primitives are investigated. If dynamic obstacles occur within the generated sweep, the robot can stop and replan.

 

Here we show the capabilities of our GPU Octree: It is possible to insert Kinect data at a 1cm resolution with 40 FPS, including the raycasting that is needed to determine the free space.

 

Collision detection for smaller objects: We use a 3D camera in the palm of a robotic hand to perform online grasp planning. GPU Voxels derives the joint angles of all fingers to grasp the object by checking collision agains the Swept-Volumes of the finger closing motions. This can be done for thousands of different grasp poses per second, so a Particle-Swarm-Optimization approach can be used to evaluate the best grasp. The video only shows the fist valid grasps per pointcloud frame, and not the optimization process.