Hi, I'm Jonathan!

I develop ways for interactive agents to perceive, understand, and adapt to our ever-changing world

I am a graduate student at the Georgia Institute of Technology in Atlanta, GA. I am currently earning my PhD in Robotics from the College of Computing. My research focus is on developing neurosymbolic methods for reinforcement learning that allow learning agents to adapt to unexpected changes in their environments. I also have collaborated with Sonia Chernova in the RAIL lab and Zsolt Kira in the RIPL lab at Georgia Tech on other ways of making machine learning and autonomy adaptable to change and novelty.

Previously I was a Robotics Engineer at Intelligent Automation, Inc. in Rockville, MD. There I focused on finding autonomous vehicle and smart device solutions under DoD research grants, specializing in creating sliding autonomy systems for teleoperators to interactively teach robots tasks and using computer vision to improve teleoperation of field robots. I earned my Masters Degree in Robotics from the University of Pennsylvania in Philadelphia, PA, and my Bachelors Degree in Physics and Mathematics from Georgetown University in Washington, D.C. For more details, check out my C.V.!

Sometimes I let this website get out of date. For the most up-to-date professional information, check out my LinkedIn.

Recent News

Just a few things that I have been up to:

  • (Mar 2022) I gave a long talk at the AAAI 2022 Spring Symposium on Designing Artificial Intelligence for Open Worlds. The talk described our research developing NovGrid environment and accompanying ontology.
  • (May 2021) Beginning my summer internship at SRI International. I will be working on using reinforcement learning to learn generative models for generating interpretable multi-agent policies based on Behavior Trees.
  • (April 2021) Our work on Semi-Supervised Continual Learning was accepted to IJCNN! We were accepted for a short oral that we will present at the virutal conference this summer.
  • (Mar 2020) I was awarded a Public Interest Technology Universities Network (PITUN) Fellowship! I will be working with Bill Sabol, a Georgia State University Criminology professor, on how machine learning can help parole officers identify which behaviors and technique reduce recidivism.
  • (July 2019) Our team won the FetchIt!: The Mobile Manipulation Challenge! with fellow collaborators in the Robot Autonomy and Interactive Learning (RAIL) lab led by Sonia Chernova at Georgia Tech we won the lab a new Fetch robot plus $7K in funds from SCHUNK.
  • (Feb 2019) Our work on Robot Tool Macguyvering was accepted to ICRA! We will present the poster at the conference in Montreal this summer.
  • (May 2018) Beginning my summer internship at Google AI in the Mobile Vision Group where I will be developing low-shot learning methods with intelligent batch construction using online importance sampling.
  • (July 2017) Attended RSS 2017 in Boston to present my work using synthetic data to improve semantic segmentation at the Workshop on New Frontiers for Deep Learning in Robotics and the Workshop on Spatial Semantic Representations in Robotics.

Highlighted Projects

images/ThailandProfile.png

[NEW] NovGrid: A Flexible Grid World for Evaluating Agent Response to Novelty

[Long Oral] AAAI 2022 Spring Symposium on Designing Artificial Intelligence for Open Worlds

A robust body of reinforcement learning techniques have been developed to solve complex sequential decision making problems. However, these methods assume that train and evaluation tasks come from similarly or identically distributed environments. This assumption does not hold in real life where small novel changes to the environment can make a previously learned policy fail or introduce simpler solutions that might never be found. To that end we explore the concept of {\em novelty}, defined in this work as the sudden change to the mechanics or properties of environment. We provide an ontology of for novelties most relevant to sequential decision making, which distinguishes between novelties that affect objects versus actions, unary properties versus non-unary relations, and the distribution of solutions to a task. We introduce NovGrid, a novelty generation framework built on MiniGrid, acting as a toolkit for rapidly developing and evaluating novelty-adaptation-enabled reinforcement learning techniques. Along with the core NovGrid we provide exemplar novelties aligned with our ontology and instantiate them as novelty templates that can be applied to many MiniGrid-compliant environments. Finally, we present a set of metrics built into our framework for the evaluation of novelty-adaptation-enabled machine-learning techniques, and show characteristics of a baseline RL model using these metrics.

images/ThailandProfile.png

Memory-Efficient Semi-Supervised Continual Learning: The World is its Own Replay Buffer

[Short Oral] 2021 International Joint Conference on Neural Networks (IJCNN)

Rehearsal is a critical component for class-incremental continual learning, yet it requires a substantial memory budget. Our work investigates whether we can significantly reduce this memory budget by leveraging unlabeled data from an agent's environment in a realistic and challenging continual learning paradigm. Specifically, we explore and formalize a novel semi-supervised continual learning (SSCL) setting, where labeled data is scarce yet non-i.i.d. unlabeled data from the agent's environment is plentiful. Importantly, data distributions in the SSCL setting are realistic and therefore reflect object class correlations between, and among, the labeled and unlabeled data distributions. We show that a strategy built on pseudo-labeling, consistency regularization, Out-of-Distribution (OoD) detection, and knowledge distillation reduces forgetting in this setting. Our approach, DistillMatch, increases performance over the state-of-the-art by no less than 8.7% average task accuracy and up to 54.5% average task accuracy in SSCL CIFAR-100 experiments. Moreover, we demonstrate that DistillMatch can save up to 0.23 stored images per processed unlabeled image compared to the next best method which only saves 0.08. Our results suggest that focusing on realistic correlated distributions is a significantly new perspective, which accentuates the importance of leveraging the world's structure as a continual learning strategy.

images/ThailandProfile.png

Tool Macgyvering: Tool Construction Using Geometric Reasoning

2019 International Conference on Robotics and Automation (ICRA)

MacGyvering is defined as creating or repairing something in an inventive or improvised way by utilizing objects that are available at hand. In this paper, we explore a subset of Macgyvering problems involving tool construction, i.e., creating tools from parts available in the environment. We formalize the overall problem domain of tool Macgyvering, introducing three levels of complexity for tool construction and substitution problems, and presenting a novel computational framework aimed at solving one level of the tool Macgyvering problem, specifically contributing a novel algorithm for tool construction based on geometric reasoning. We validate our approach by constructing three tools using a 7-DOF robot arm.

Unbiasing Semantic Segmentation For Robot Perception using Synthetic Data

RSS 2017 Workshop on New Frontiers for Deep Learning

The ability of a robot to reason about the geometry and semantics of its environment is fundamental to interactive robot behaviors, but often challenging due to perception frameworks that are trained on too little data or data not representative of the robots environment. In this work, we investigate the potential gains of using synthetic data to augment the training process of convolutional neural networks designed to enable real-time semantic segmentation for robots with limited real-world training data. We investigate the degree to which a larger amounts of data improves performance when training such a model, the relationship between the way a deep neural network is trained using multiple sources of synthetic segmentation data to pretrain standard segmentation datasets that apply to robotics and autonomous driving, and show that our method outperforms both training from scratch and standard data augmentation practices like pretraining on ImageNet. We show that synthetic data does continue to improve these models in spite of real-time model architectures having many fewer parameters than typical deep neural networks, and therefore hypothetically less representational power. Finally, we show how this approach generalizes to small purpose-built robot vision datasets on data acquired using an HRI robot.

Contact Me

If you have any questions for me please feel free to e-mail me at balloch@gatech.edu.