Andrew Spielberg
AI-Powered Design And Manufacturing For Embodied AI
Postdoctoral Fellow
Harvard University
Mechanical Engineering & Materials Science
Email:
aespielberg@seas.harvard.edu
Google Scholar:
https://scholar.google.com/citations?user=8JeQMMUAAAAJ
Morphological Intelligence
Meets Artificial Intelligence
Meets Artificial Intelligence
My mission is to enable anyone to be able to design functional artifacts across scales and domains, and with a special emphasis on robotics and other cyberphysical machines. I want to empower novices and accelerate experts' workflows.
I research simulation methods, design algorithms, digital manufacturing processes, and methods for overcoming the sim-to-real gap, for inventing in both virtual and physical worlds. With the right tools, we can augment human creativity, advance more capable computer creativity, and build computational systems that work for everybody, not just a select few.
My work has been published at top venues and featured in popular science periodicals such as, well, Popular Science, and others like Scientific American, TechCrunch, and more. I am a recipient of the Unity Global Fellowship, the DARPA I2O Fellowship, the MIT Sandbox Innovation Fund, and a Harvard GRID $100K award.
Presently, I am a Postdoctoral Associate at Harvard University, where I work with Prof. Jennifer A. Lewis in the aptly named Lewis Lab and collaborate closely with Prof. Karen Liu at Stanford. I received my PhD from MIT's Computer Science and Artificial Intelligence Lab, where I was advised by Daniela Rus and Wojciech Matusik. During my PhD, I did two stints at Disney Research, Pittsburgh and Zürich. In Olden Days of Yore, I did my undergrad/Master's at Cornell.
I am on the academic job market and will be applying FA '23.
Feel free to get in touch!
Updates
(✈️ = Spiel's On Wheels, i.e. travel)
11/17/2023 -- Our new paper on differentiable visual computing is live, in Nature Machine Intelligence!
11/10/2023 -- Time Crystal has been accepted to Nature Futures. More details when it's in print!
✈️ 10/21/2023 -- Thank you to Maker Faire: Bay Area for hosting me for a talk about The Scion, and for helping to sell early editions!
10/17/2023 -- I finally updated my old forgotten website with a new website I'm sure to forget to update. Please bear with me as some content is still being added and some formatting may be adjusted.
Funding and Collaborations
Research Funders (They like the shoutout)
Industry Collaborators
Highlighted Research Projects
For all core research projects, please see the publications page.
Research Questions
Generative yet optimal, fabricable, certifiable robot co-design
User-Centric AI-Powered Co-Design Algorithms
How do we enable creative, fast, yet (pareto-)optimal design of robots and other devices across domains and application spaces?
Data for physical designs will always be sparse in new regimes. How can AI design in these data-sparse or data-free regimes, and provide certificates for design performance?
What are the the necessary simulation and modeling tools needed to power those algorithms?
Digital Fabrication Powered By Physical Intelligence
The world reacts to stimuli; for intelligent matter to sense, it just needs to measure those reactions. How can we program matter to embed and extract information about the world?
How do we embed sensing and actuation and control for material properties and geometry all at once, in end-to-end digital manufacturing processes?
How do we realize programmable physical devices across scales, with as little human intervention as possible in the fabrication process?
Modeling For Design across Domains and Morphologies
When someone invents a novel device in the virtual world, it should just work in the physical world. How do we overcome the sim-to-real gap, not just for one robot, but for any robot?
How can we quantify and control uncertainty in design and manufacturing?
How do we marry analytical simulation with data-driven models, providing the best of both worlds?
This last part "closes the loop," allowing data from designs to improve and adapt modeling (and future designs) to novel domains.
Vision
I take systems-, algorithmic-, user-, and society-centric approaches to research for impact.
▲ Building modular design technology stacks.
🤖 Methods that learn from the data they produce:
i.e. supervise on the simulator, not an external dataset.
🖱️ Keep the practitioner's wants and needs in mind:
Research is most useful when it is usable!
🤓 Make core research accessible through open-sourcing/
releasing what we can.