CS326: Advanced Topics in Robotic Manipulation
Proposed NERF-based perception system and interface with contact-implicit trajectory optimization for manipulation in novel, contact-rich environments.

AA277: Multi-Robot Systems
Investigated modifications to nonlinear pose graph optimization (ala SE-Sync) to increase robustness to adversarial agents via alternative noise models and graph pruning.

Convex Optimization I and II
Developed ADMM-based solver for L2 fuel-optimal spacecraft control.

AA 203: Optimal Control
For the final project I developed a floating-body dynamics model for a hopping robot and used discrete nonlinear optimization to find hopping trajectories that minimized the actuator efforts. While many controllers for legged robotics optimize using a fixed gait pattern, I instead allowed the solver to find the optimal timing and location for the foot steps by using nonlinear complementarity constraints.


Figure AA203.1 Optimal hopping trajectory for a single-legged robot.


Figure AA203.2 Dynamics model for single-legged robot.

CS238: Decision Making Under Uncertainty
For the final project of Decision Making Under Uncertainty, I worked with a partner to use reinforcement learning to teach a quadruped how to do backflips. We posed the dynamics and state estimation of Stanford Doggo, the quadruped my team developed from the ground-up, as a POMDP and we solved for the optimal control using Proximal Policy Optimization.


Figure CS238.1 Stanford Doggo


Figure CS238.2 Learned backflip


CEE 32H: Responsive Structures
I worked with a small group of students to ideate, prototype, and create what we call "Swell," an immersive sculpture constructed out of formable plywood panels.



ME 203: Design and Manufacturing

In ME203, I designed, prototyped, and fabricated a bike light that uses your phone's flash led as the light source.

Processes
Milling, Turning, Laser cutting, Bead blasting, Anodizing, Finishing, CAD



nathankau.com
Palo Alto, California