Zhen Peng

Alumni of the Research Group Sensorimotor Learning and Decision-Making

Forschungsinteressen

I am a third year PhD student at the Max-Planck Institutes for Biological Cybernetics and Intelligent Systems, also a member of the Sensorimotor Learning and Decision-Making Group headed by Daniel A. Braun. My research focuses on developmental robotics, computational neuroscience, intrinsic motivations and motion complexity.

Project A: Assessing randomness and complexity in human motion trajectories through analysis of symbolic sequences

Complexity is a hallmark of intelligent behavior consisting both of regular patterns and random variation. To quantitatively assess the complexity and randomness of human motion, we designed a motor task in which we translated subjects' motion trajectories into strings of symbol sequences. In the first part of the experiment participants were asked to perform self-paced movements to create repetitive patterns, copy pre-specified letter sequences, and generate random movements. To investigate whether the degree of randomness can be manipulated, in the second part of the experiment participants were asked to perform unpredictable movements in the context of a pursuit game, where they received feedback from an online Bayesian predictor guessing their next move. We analyzed symbol sequences representing subjects' motion trajectories with five common complexity measures: predictability, compressibility, approximate entropy, Lempel-Ziv complexity, as well as effective measure complexity. We found that subjects’ self-created patterns were the most complex, followed by drawing movements of letters and self-paced random motion. We also found that participants could change the randomness of their behavior depending on context and feedback. Our results suggest that humans can adjust both complexity and regularity in different movement types and contexts and that this can be assessed with information-theoretic measures of the symbolic sequences generated from movement trajectories.

Project B: Curiosity-driven learning with Context Tree Weighting

Research on intrinsic motivation has emerged as an important new field in robotics addressing the question of how intelligent behavior might develop through processes of self-organization. Many traditional learning and control schemes typically assume predefined utility or cost functions that are externally imposed to achieve very specific behaviors. In contrast, intrinsic rewards do not specify any particular behaviors, but allow the agent to explore the world (including its own body) in a developmental fashion. Several measures have been proposed as possible intrinsic rewards to drive learning agents – including compression progress, learning progress, predictive information and empowerment. All these four intrinsic motivation functions have showed promising and interesting learning behaviors. In this pilot study, we created a very simplified environment where the agent can execute three different actions that lead respectively to a constant deterministic response from the environment, a regular response requiring memory and a random feedback.  We implemented a learning agent based on Context Tree Weighting (CTW), which is a generic Bayesian binary sequence prediction algorithm, to learn the probability distribution of perceptions conditional on the agent’s actions. Two intrinsic motivations have been investigated in our study: learning progress and compression pregress. In both scenarios, our agent learns first the action that is easiest to learn, then switches to the next action once it gets bored and can predict the consequences. Then the agent switches to other actions that still allow for learning. Our preliminary results suggest that Context Tree Weighting might provide a very general representation that is useful to study problems of development.

Vita

EDUCATION

  • 11/2012 – now        Phd Student at the Max Planck Institute for Biological Cybernetics & Intelligent Systems.
  • 10/2006 – 10/2012  Diplom in Computer Science at University of Stuttgart, Germany

EXPERIENCE

  • 11/2011 – 05/2012  Project "NavOScan" at Innovation Center Iceland (ICI) in Reykjavik, Iceland
  • 04/2011 – 10/2011   Project "ASBUS" at the University Institute for Visualization and Interactive Systems (VIS) at the University of Stuttgart, Germany
  • 11/2010 – 05/2011 Development of a SLAM Framework at the Fraunhofer Institute for Manufacturing Engineering and Automation (IPA) in Stuttgart, Germany
  • 07/2009 – 10/2012   Project "ASBUS" at the University Institute for Visualization and Interactive Systems (VIS) at the University of Stuttgart, Germany
  • 03/2008 – 03/2009  Project Assistence at the University Institute of Architecture of Application Systems (IAAS) at the University of Stuttgart, Germany
Zur Redakteursansicht