Real-Time Person and Object Detection

The following video shows a real-time, GPU-based RGBD person detector. It is based on a Deformable Parts Model (DPM) with a joint RGB+Depth feature descriptor to perform high-accuracy person detection at 5Hz while requiring less than 10 Watts on a single 2014 consumer-grade embedded chip. More details can be found in the IROS 2015 paper "Depth-Augmented Deformable Parts Models for RGBD Person Detection on Embedded GPUs".

The following video demonstrates a system for real-time, GPU-accelerated multi-class object detection using a sparse-coded deformable parts model (DPM). It is based on the approach in the ECCV 2012 paper "Sparselet Models for Efficient Multiclass Object Detection".

Physics-Based Planning

My PhD research was about physics-based motion planning. The following videos show some interesting planning examples for several challenging domains. For more information about this work, please have a look at my PhD Thesis section.

Simulated Domains: Minigolf, Soccer, Pool, Many-Dice

Robot Minigolf Part 1: Configuration

Robot Minigolf Part 2: More Examples

Physics-Based Dribbling


RoboCup is an international robot soccer competition. Our team "CMDragons" participates in the Small Size League where robots play 5 vs. 5 soccer games with a golf-ball.

CMDragons Highlights from RoboCup 2008, Suzhou, China

CMDragons Highlights from RoboCup 2007, Atlanta

CMDragons on National Television!

RoboCup 2009 - Kicker Strength Shootout

Note that this "competition" is not officially part of RoboCup... but it was fun nevertheless!

Computer Animation

The following videos were generated by simulators I wrote for homework assignments of Carnegie Mellon's "Physical Simulation for Computer Animation" course, taught by Adam Bargteil in Fall 2007.

Invertible Finite Elements Simulation

Smoke Simulator

Behavioral Particle System

Structured Light

The following video shows a light striping system I developed in 2007 using a simple off-the-shelf projector and single firewire camera. Using the offset between camera and light-source, it is possible to reconstruct the 3D shape of objects through triangulation.