Robotic Arm + ACT Policy
Building a robotic arm and training an action-chunking-with-transformers (ACT) policy on it using Hugging Face / LeRobot. Still a work in progress.
- Demo
- x.com/lucjbeck
Student at the University of PennsylvaniaHOME // 00
Luc Beck
Studying Electrical Engineering and Artificial Intelligence at Penn. Working on the intersection of Physics and AI.
Dual pick-and-place robotic arms, reconstructing the Penn coat of arms.
About
Premise
Take a system apart and put it back together until you understand exactly how it works. Then build the version that should exist.
My interests sit where hard physics meets computational theory. In my scientific research, that means biophysics: I built mathematical models of how DNA-based polymers can be used to perform experiments (aka 'lab on a molecule'), which fed into a drug-discovery platform now developing therapeutics. Outside scientific research, I build systems that move and learn, like a robotic arm I built and taught to manipulate objects via imitation learning.
I'm also interested in societal problem solving through serial entrepreneurship.
Education
- Institution
- University of PennsylvaniaSchool of Engineering and Applied Science
- Degree
- B.S.E., Electrical Engineering and Artificial IntelligenceDual major
- GPA
- 3.95 / 4.00
- Class of
- 2029
- Coursework
- Computer ScienceObject-Oriented Programming, Scientific ComputingMathematicsLinear Algebra, Multivariable Calculus, Discrete MathematicsPhysicsMechanics, ElectromagnetismEconomicsMicroeconomics, Macroeconomics
- Honors
- U.S. Presidential ScholarOne of up to 161 students named nationally each year
Engineering
Next: Robotic Hand
Building a dexterous robotic hand paired with an instrumented glove. The glove captures human hand motion so the robot can be trained directly on teleoperated demonstrations, giving it a path to develop the fine-grained manipulation that standard robot arms lack.
Research
DNA Nanoswitches for Drug Discovery
Three summers building a DNA-nanoswitch platform for therapeutic discovery. I derived polymer-physics and enzyme-kinetic models predicting nanoswitch conformational states, reaction rates, and binding equilibria, and used AlphaFold and Python to model nanobody-target interactions across autoimmune and viral pathways. I ran native PAGE gel-shift assays measuring binding enhancement of nanoswitch-discovered nanobodies on the PD-1/PD-L1 complex implicated in autoimmune disease. The platform has since spun into a company developing therapeutics, with candidates advancing toward clinical trials.
Publication
Accepted — Awaiting Publication
Tethered Molecular Interactions in DNA Nanoswitches: Modeling and Controlling Local Concentration. With Wesley Wong (Harvard Medical School) and postdoctoral researcher Andy Ward.
This paper investigates how spatial constraints influence molecular binding by using DNA nanoswitches, programmable DNA scaffolds that position binding partners at defined distances. By integrating gel-based experiments with a polymer physics model, we quantify how binding-site arrangement affects the probability of loop formation through changes in effective local concentration. The results provide a generalizable approach for predicting and controlling tethered biomolecular interactions in synthetic and biological systems.
Experience
Summer 2026
Washington, DC
Construct Capital
Summer Analyst
I'm building AI agent infrastructure that streamlines the firm's internal workflows and the screening of inbound companies, accelerating deals through the pipeline.
Summers 2024 and 2025
Cambridge, MA
Northpond Ventures
Venture Capital Intern
I evaluated spinning DNA-nanoswitch drug-discovery technology out into a new company, Merge Bio, assessing platform scalability, IP positioning, and commercialization strategy. I built the financial model end to end, a risk-adjusted revenue build, DCF, and sensitivity analyses spanning market sizing, pricing benchmarks, and probability-of-success adjustments across clinical phases. I also authored the scientific positioning on nanoswitch-based therapies for Type 1 Diabetes, covering PD-1/PD-L1 agonist mechanisms and the competitive landscape. The work fed into a $15M seed round.