Joseph Campbell

Postdoctoral Researcher @ CMU

This site is no longer actively maintained!

Please see my new website here: joe-campbell.github.io/website/

About Me

My research area is the intersection between robotics and machine learning, with a focus on explainable models for human-robot interaction.

More specifically, my dissertation work focuses on the spatiotemporal aspects of inference for a collaborative, physical task given partial observations of the human with probabilistic methods.

I am interested in many aspects of artificial intelligence and machine learning, having worked on a wide range of projects ranging from workload prediction algorithms for embedded GPUs to physical interaction with human-scale robots to traffic light classification in fully autonomous vehicles. I am always open to new collaborations!

***

I am currently a postdoctoral researcher at Carnegie Mellon University in the Robotics Institute.

I completed my PhD in the Interactive Robotics Lab at Arizona State University.

My CV can be found HERE, LinkedIn HERE, and I can be reached at jacampb1@asu.edu.

I maintain a library for Bayesian Interaction Primitives which can be found on Github.

Latest Blog Updates

ICRA 2020 moved to virtual conference; presentation posted below - 05/23/2020

Two papers accepted to ICRA 2020 - 02/03/2020

I will be presenting my work at IROS 2019 - 10/28/2019

Invited talk at IROS 2019 - 10/25/2019

IntPrim v2.0 Released on Github! - 10/20/2019

(read more)

Research

Below is a select sample of some of the research projects that I have worked on and the most relevant publication.

Whole-Body Human-Robot Haptic Interaction

Language-Conditioned Imitation Learning

campbell_icra2020_video_final.mp4

Whole-body human-robot interactions rely on non-traditional sensing modalities, such as haptic sensing. However, these sensors typically exhibit properties such as spatiotemporal locality, which make utilizing them in learning algorithms quite challenging. We present a method for employing haptic sensing to learn whole-body interactions in a hugging scenario with a human partner, utilizing Bayesian Interaction Primitives.

Presentation available below.

Learning Whole-Body Human-Robot Haptic Interaction in Social Contexts, J. Campbell, and K. Yamane, ICRA 2020

Most imitation learning approaches involve extracting policy parameters from a single modality of data, e.g., visual data. No adequate communication channel exists between the human expert and the robot to describe critical aspects of the task, such as the properties of the target object or the intended shape of the motion. We introduce a method for incorporating unstructured natural language into imitation learning in order to resolve ambiguity and inflexibility in goal specifications.

Language-Conditioned Imitation Learning for Robot Manipulation Tasks, S. Stepputtis, J. Campbell, M. Phielipp, S. Lee, C. Baral, H. Ben Amor , NeurIPS 2020

Probabilistic Multimodal Modeling for HRI

HRI With Musculoskeletal Robots

Bayesian Interaction Primitives suffer from linearization errors and assumptions made about the parametric family of the prior/posterior probability distributions, which are especially relevant in multimodal scenarios. We introduce ensemble Bayesian Interaction Primitives (library here), a Monte Carlo-based approach which is less susceptible to these error sources, yielding improved inference accuracy and computational complexity.

Presentation available below.

Probabilistic Multimodal Modeling for Human-Robot Interaction Tasks, J. Campbell, S. Stepputtis, and H. Ben Amor, RSS 2019

Musculoskeletal robots that are based on pneumatic actuation have a variety of properties, such as compliance and back-drivability, that render them particularly appealing for human-robot collaboration. However, programming interactive and responsive behaviors for such systems is extremely challenging due to the nonlinearity and uncertainty inherent to their control. We propose an approach for learning Bayesian Interaction Primitives for musculoskeletal robots given a limited set of example demonstrations.

Learning Interactive Behaviors for Musculoskeletal Robots Using Bayesian Interaction Primitives, J. Campbell, A. Hitzmann, S. Stepputtis, S. Ikemoto, K. Hosoda, and H. Ben Amor, IROS 2019

SLAM and Learning from Demonstration

Foldable Robotics and Bio-inspiration

Simultaneous localization and mapping techniques are applied to the Bayesian inference problem found in learning from demonstration, resulting in a new framework: Bayesian Interaction Primitives (library here). This results in improved performance and inference accuracy when compared to an existing state-of-the-art method.

Presentation available below.

Bayesian Interaction Primitives: A SLAM Approach to Human-Robot Interaction, J. Campbell, H. Ben Amor, CoRL 2017

Foldable robots constructed from laminate materials are useful for scenarios which call for inexpensive, easily fabricated robots. Using landmine detection in desert environments as a motivating example, we built the C-Turtle (more info here) which takes inspiration from its biological counterpart and utilizes reinforcement learning to efficiently navigate through granular media.

From the Lab to the Desert: Fast Prototyping and Learning of Robot Locomotion, K.S. Luck, J. Campbell, M.A. Jansen, D.M. Aukes, and H. Ben Amor, RSS 2017

GPU Performance Prediction

Traffic Light Detection

Dynamic Voltage Frequency Scaling, commonly known as throttling, is tricky to employ on GPUs due to the sensitivity of application performance to clock frequency. Here we constructed a light-weight prediction algorithm to estimate frame time sensitivity to GPU frequency in real-time, with the goal of enabling power-efficient operation without sacrificing application performance.

Adaptive Performance Prediction for Integrated GPUs, U. Gupta, J. Campbell, U.Y. Ogras, et al, ICCAD 2016.

Traffic light detection in autonomous vehicles is typically based on image detection techniques driven by optical cameras. However, these methods are susceptible to occlusion, poor lighting, and other issues. In this project, we performed complementary traffic light detection by analyzing the movement patterns of nearby vehicles.

Traffic Light Status Detection Using Movement Patterns of Vehicles, J. Campbell, H. Ben Amor, M.H. Ang, and G. Fainekos, ITSC 2016.

Modeling Cooperation between Autonomous Vehicles

Multi-robot Path Planning

Safety analysis of cooperation between autonomous vehicles often requires verification of the underlying protocols to ensure constraints are respected and fail states are avoided. We introduce a new verification framework that abstracts the communication through π-calculus and the underlying vehicle dynamics with hybrid automata, allowing for analysis of heterogeneous vehicle formations.

Modeling Concurrency and Reconfiguration in Vehicular Systems: A π-calculus Approach, J. Campbell, C.E. Tuncali, et al, CASE 2016.

Introduced in previous work, DisCoF is a cooperative path planning algorithm for distributed robotic systems featuring limited sensing and communication range. This extension improves the performance of DisCoF by introducing asynchronous planning and the ability for coupled robots to decouple based on appropriate heuristics.

DisCoF+: Asynchronous DisCoF with Flexible Decoupling for Cooperative Pathfinding in Distributed Systems, K. Kim, J. Campbell, W. Duong, Y. Zhang, and G. Fainekos, CASE 2015.

Publications

A list of my peer-reviewed conference, journal, and workshop publications can be found below.

S. Stepputtis, J. Campbell, M. Phielipp, S. Lee, C. Baral, and H. Ben Amor. Language-Conditioned Imitation Learning for Robot Manipulation Tasks. Conference on Neural Information Processing Systems (NeurIPS), Virtual, December 2020. Spotlight (top ~4% submitted papers).

G. Clark, J. Campbell, and H. Ben Amor. Learning Predictive Models for Ergonomic Control of Prosthetic Devices. Conference on Robot Learning (CoRL), Virtual, November 2020.

J. Campbell and K. Yamane. Learning Whole-Body Human-Robot Haptic Interaction in Social Contexts. International Conference on Robotics and Automation (ICRA), Paris, France, May 2020.

G. Clark, J. Campbell, S.M.R. Sorkhabadi, W. Zhang, and H. Ben Amor. Predictive Modeling of Periodic Behavior for Human-Robot Symbiotic Walking. International Conference on Robotics and Automation (ICRA), Paris, France, May 2020.

S. Stepputtis, J. Campbell, M. Phielipp, C. Baral, and H. Ben Amor. Imitation Learning of Robot Policies by Combining Language, Vision and Demonstration. NeurIPS Workshop on Robot Learning (NeurIPS-WRL), Vancouver, Canada, December 2019.

K. Bagewadi, J. Campbell, and H. Ben Amor. Multimodal Dataset of Human-Robot Hugging Interaction. AAAI Fall Symposium on Artificial Intelligence for Human-Robot Interaction (AI-HRI), Arlington, Virginia, November 2019.

J. Campbell, A. Hitzmann, S. Stepputtis, S. Ikemoto, K. Hosoda, and H. Ben Amor. Learning Interactive Behaviors for Musculoskeletal Robots Using Bayesian Interaction Primitives. IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Macau, China, November 2019.

J. Campbell, S. Stepputtis, and H. Ben Amor. Probabilistic Multimodal Modeling for Human-Robot Interaction Tasks. Robotics: Science and Systems (RSS), Freiburg, Germany, June 2019.

J. Campbell and H. Ben Amor. Bayesian Interaction Primitives: A SLAM Approach to Human-Robot Interaction. Conference on Robot Learning (CoRL), Mountain View, California, November 2017.

M.A. Jansen, K.S. Luck, J. Campbell, H. Ben Amor, and D.M. Aukes. Bio-inspired Robot Design Considering Load-bearing and Kinematic Ontogeny of Cheloniodea Sea Turtles. Conference on Biomimetic and Biohybrid Systems (Living Machines), Stanford, California, July 2017.

K.S. Luck*, J. Campbell*, M.A. Jansen*, D.M. Aukes, and H. Ben Amor. From the Lab to the Desert: Fast Prototyping and Learning of Robot Locomotion. Robotics: Science and Systems (RSS), Boston, Massachusetts, July 2017. Note: first three authors have shared first-authorship.

J. Campbell, H. Ben Amor, M.H. Ang Jr., and G. Fainekos. Traffic Light Status Detection Using Movement Patterns of Vehicles. IEEE International Conference on Intelligent Transportation Systems (ITSC), Rio de Janeiro, Brazil, November 2016.

U. Gupta, J. Campbell, U.Y. Ogras, R. Ayoub, M. Kishinevsky, F. Paterna, and S. Gumussoy. Adaptive Performance Prediction for Integrated GPUs. IEEE/ACM International Conference on Computer Aided Design (ICCAD), Austin, Texas, November 2016.

J. Campbell, C.E. Tuncali, P. Liu, T.P. Pavlic, U. Ozguner, and G. Fainekos. Modeling Concurrency and Reconfiguration in Vehicular Systems: A Pi-Calculus Approach. IEEE International Conference on Automation Science and Engineering (CASE), Fort Worth, Texas, August 2016.

J. Campbell, C.E. Tuncali, T.P. Pavlic, and G. Fainekos. Toward Modeling Concurrency and Reconfiguration in Vehicular Systems. Interaction and Concurrency Experience Satellite Workshop of DisCoTec (ICE), Heraklion, Greece, June 2016.

K. Kim, J. Campbell, W. Duong, Y. Zhang, and G. Fainekos. DisCoF+: Asynchronous DisCoF with Flexible Decoupling for Cooperative Pathfinding in Distributed Systems. IEEE International Conference on Automation Science and Engineering (CASE), Gothenburg, Sweden, August 2015.

Talks

Some of my talks may be recorded during the course of live streaming at conferences. I will link them here when available.

ICRA 2020

campbell_icra2020_presentation.mp4

RSS 2019

https://youtu.be/vgkxR9TDqhY?t=9913 (if embedding doesn't work)

CoRL 2017

https://youtu.be/_9Ny2ghjwuY?t=26862 (if embedding doesn't work)

Press Releases

My research has been featured in a number of press releases. Below is a select list although by no means comprehensive.

Gizmodo: Half My Nightmares Start With This Giant Cyborg Teddy Bear Learning How to Hug

CBS News: Arizona researchers developing robots that can learn

BBC News: C-Turtle: The landmine-detecting robot 'turtle'

RaspberryPi: Landmine-clearing Pi-powered C-Turtle

ASU Now: ASU-designed C-Turtle robot teaches itself to get around

New Scientist: Robotic turtles can be used to detect landmines in the desert

Bio

I am a graduate student at Arizona State University pursuing a PhD degree in Computer Science. My research focus is Bayesian inference applied to physical human-robot interaction. Prior to this, I earned a BS in Computer Science and MS in Computer Engineering from Arizona State University. Outside of academia, I have more than six years of experience as a Software Engineer at companies such as Garmin and Intel.

In my spare time, I enjoy traveling, flailing around on a guitar, reading sci-fi books, and convincing myself that I'm not a total couch-potato by taking mild hikes and bicycle rides.

Report abuse