adamsetapen.com

home > academics

.: academics :.

I am a computer scientist and roboticist with a strong interest in robotic learning, especially through human-robot interaction. I'm currently a graduate student in the Personal Robots Group at the MIT Media Lab. My research focuses on finding interaction patterns and adapting a robot’s performance through crowdsourcing. I’m also interested in finding novel ways to harness humans for transferring knowledge to artificial agents. A common thread in my work is investigating effective routes towards robotic learning that facilitates a symbiotic human-robot relationship.

publications

W. Bradley Knox, Adam Setapen, and Peter Stone. Reinforcement Learning with Human Feedback in Mountain Car. To appear, AAAI 2011 Spring Symposium - Help Me Help You: Bridging the Gaps in Human-Agent Collaboration., Palo Alto, CA - March 2011.
Download: [pdf] (584kB)

Adam Setapen, Michael Quinlan, and Peter Stone. Beyond Teleoperation: Exploiting Human Motor Skills with MARIOnET. In AAMAS 2010 Workshop on Agents Learning Interactively from Human Teachers (ALIHT), Toronto, Canada - May 2010.
Supplemental video cited in the paper.
BibTeX Download: [pdf] (1.8MB)

Adam Setapen, Michael Quinlan, and Peter Stone. MARIOnET: Motion Acquisition for Robots through Iterative Online Evaluative Training (Extended Abstract). In The Ninth International Conference on Autonomous Agents and Multiagent Systems (AAMAS), International Foundation for Autonomous Agents and Multiagent Systems, Richland, SC, May 2010.
Supplemental video cited in the paper.
BibTeX Download: [pdf] (332kB)

Adam Setapen. Exploiting Human Motor Skills for Training Bipedal Robots. Undergraduate Honors Thesis/Technical Report HR-09-02. Committee: Peter Stone (chair), Dana Ballard, Gordon Novak.
BibTeX Download: [pdf] (2.3MB)

undergraduate honors thesis

Title: Exploiting Human Motor Skills for Training Bipedal Robots

Abstract: Although machine learning, reinforcement learning, and learning from demonstration have improved the rate and accuracy at which robots can gain intelligence from humans, they haven't reached the rapid rate at which humans are able to acquire new knowledge. Many systems that exploit imitation learning use simple positive and negative reinforcement, and place the burden of learning completely on the computer. This neglects the expressive capabilities of humans, as well as their remarkable ability to quickly refine motor skills. While passive dynamics offers the most human-like locomotion for bipedal robots, it also relies on particular design specifications. This thesis presents a general Framework for Interactive Control of a Humanoid by Motion Capture (FICHMC), that offers rapid motion development for large classes of bipedal robots. Essentially, a human in a motion-capture laboratory "puppets" a biped, with a real-time mapping from human to robot. The training process requires no technical knowledge and provides a natural interface for humans to directly transfer skills to robots.

Complete Paper: PDF

Video: Hi-Res | Low-Res

interesting projects

My site for Neil Gershenfeld's epic class, How to Make (almost) Anything.

A bootable x86 operating system I developed with my good friend Jose Falcon for my undergraduate operating systems course.

  • Language: C
  • Requirements: x86 processor, or an emulator (such as bochs)
  • Download

A pipelined processor designed and implemented (using an extended version of the LC-3 architecture) for my undergraduate computer architecture course with Daniel Chimene.

  • Language: Verilog
  • Requirements: Verilog simulator (such as VCS)
  • Download

A genetic algorithm for learning to play keepaway in the robotic soccer domain. I designed this algorithm in my undergraduate course, Autonomous Multiagent Systems.

Simultaneous Localization and Mapping (SLAM) simulator designed in my graduate robotics course. Displays a map of a mobile robot's probabilistic position in its environment using a particle filter.

A monadic parser, typechecker, and evaluator for a simply-typed lambda calculus augmented with booleans, natural numbers, fix-points and references.

An implementation of the RSA encryption/decription protocol.