- Autonomous quadcopter using monocular SLAM
- Invertible representations for speech recognition and synthesis
- Improved multi-touch gesture recognition algorithms for Tacchi.
- Autonomous quadcopters and neural headsets
- Multi-touch gesture recognition for the Tacchi Table(1st place in EOH 2011)
- Knowledge Extractor (presented at EOH)
- Networked Inference Engine (unimplemented idea)
- Instant Messenger Bot (presented at EOH)
- Stock Price Predictor (interested? form a team)
We worked on a traffic simulator comparing greedy and global algorithms for automated road traffic control. The greedy algorithm was intended to simulate real-world traffic situations; each vehicle simply adopted the shortest path to its destination. The global algorithm took into account regional variations in traffic density, and on average generated quicker routes. EOH poster.
The Lifelong Planning project was an effort to optimize STRIPS planning by generating large numbers of plans in a particular domain, and then combining commonly occuring sequences of operators into 'super-operators' that could be used in future plans.
SIGART worked side-by-side with SIGSOFT on the DEUS project. Its goal was to create realistic interactive storylines in real time. SIGART's role in this project was implementation of the so-called "director," a component whose task was to ensure that the plotline progressed according to a rough outline defined by a human author. This was accomplished through the use of planning algorithms, state-space search techniques, and other Artificial Intelligence approaches to the problem.