- Research
Research Projects
Intent Recognition (thesis)
The ability to understand the intentions of others, and to use that understanding to act appropriately, is important in many areas of robotics. This ability, known as intent recogntion is used in areas including the development of socially aware robots, robots which are able to detect undesireable behaviors before they happen, and even intelligent prosthetics which can use predictions of intent to perform appropriate actions. In this project, we address some problems with existing methods for intent recognition. First, we scaled an existing method for single-agent intent recognition to work efficiently in scenes involving large numbers of agents. Once this was working, we were able to use the recognized single-agent intentions as inputs to an activation network in which multi-agent intentions (intentions which require the cooperation of two or more agents) were encoded. The details of both of these projects can be found here.Source code for the intent recogntion algorithms, as well as a simulation environment which works with them, can be accessed via git:
- Single-agent intent recognition:
- git clone https://github.com/dbigelow/intent.git
- Multi-agent intent recognition:
- git clone https://github.com/dbigelow/activation_network.git
- Communication packages:
- git clone https://github.com/dbigelow/commlink.git
- git clone https://github.com/dbigelow/vis.git
- Simulation environment:
- git clone https://github.com/dbigelow/shipSim.git
Mapping and Exploration Using Expansive Spaces Trees (unpublished)
In this project, a method for exploring an unknown environment was developed using concepts from path planning. The expansive spaces tree (EST) method for path planning was adapted for use in frontier-based exploration, both with a single agent, and with multiple agents. This algorithm allowed us to guide the exploration by simply varying the probability distribution function over the nodes in the EST. This ability to easily change the manner in which the agents explored their environment allows us to use the same exploration algorithm for different applications, from exploring as rapidly as possible, to exploring in a way which allowes efficient travel in the future. A paper detailing the single agent algorithm can be found here, and the extension to the multi-agent algorithm can be found here.Source code for both projects can be accessed through git (coming soon):
- Single agent:
- Multi agent: