Elastic rods in joint coordinates, for robot simulators2025–2026 · manuscript in preparation, sole author

• Derived one swing–twist rod energy with an exact, closed-form ball-joint torque, and implemented it four ways (NumPy, JAX, PyTorch inside mjlab, a C++ plugin for MuJoCo) that agree to 2×10⁻¹¹. The chain converges to Euler–Bernoulli beam theory at fourth order.
• Found that MuJoCo's stock cable torque is not the gradient of an energy (off by 7.7–46.7% once the bend axis tilts) and that its RK4 is second order, not fourth, for non-planar rotations.
• Control on the verified model: a whip crack optimized by reverse-mode differentiation through MJX (tip at 68 m/s, 5.8 mm from its target); a periodic LQR that holds an unstable cracking orbit; PPO in 4,096 parallel environments (MuJoCo Warp) that rears a 16-link chain; a receding-horizon controller that rights a tumbling glider in 10 of 10 launches.

Entangled rod packings: optimization, contact engines, X-ray CT2021–2026 · PNAS 2025 · PRL under revision

• Wrote a JAX optimizer (FIRE, on GPU) that maximizes the entanglement of up to 2,000 rods without overlap, and a C++17 engine for rigid rods in contact (cone complementarity by projected Gauss–Seidel, soft contact, OpenMP, a CUDA broad phase).
• Showed that maximally entangled packings cage themselves and hold together by friction alone. Re-ran the dynamics in MuJoCo to test which results survive a change of contact model: the cooling law does (time constant within 15%), contact lifetimes do not.
• Measured real packings by X-ray micro-CT, with a pipeline that traces about 4,600 rods in a 2000 × 2000 × 741 voxel scan in 12 minutes.

Shape-morphing lattices: prediction and inverse design2024–2026 · PNAS 2026

• Discrete elastic rod simulations that predict how rotationally 3D-printed lattices morph when heated (with the Lewis group at Harvard).
• Inverse design in JAX: a damped sparse Newton solver with stability checks follows the equilibrium out of the flat state, and Gauss–Newton fits the design to a target surface.

Calibrating a simulator against hardware2026 · with experimental collaborators

• A simulator for chains of centimeter-scale vibration-driven robots, with hard and soft constraints. Rotational noise is fitted from tracking of single robots; the calibrated model matches 23 of 30 cells of the measured stability map. I rejected a variant that fitted better but stretched the joints beyond their measured play.

Research Associate in Applied Mathematics 2026–present

Postdoctoral Fellow 2021–2026

Harvard John A. Paulson School of Engineering and Applied Sciences, Cambridge, MA · with L. Mahadevan

• Simulation, optimization and experiments on contact-rich filamentary systems: collision handling, nonlinear rod mechanics, reduced-order models, inverse design.
• Worked with the Harvard Microrobotics Laboratory on grasping by entanglement (PNAS 2022) and with the Lewis group on printed shape-morphing lattices (PNAS 2026).

Postdoctoral Fellow 2020–2021

Rowland Institute at Harvard, Cambridge, MA

• Modeling, fabrication and flow measurement (particle image velocimetry) of shark-skin-inspired drag-reducing surfaces.

Postdoctoral Fellow 2019–2020

Seoul National University, Seoul, South Korea

• Models and experiments for ionic-hydrogel actuators and soft electroreceptors (Science Robotics 2020, Science Advances 2021), and the mechanics of mud nests (PNAS 2021).