Research

Research that meets the real world.

Learning, estimation, planning, and control for robots and autonomous systems.

My work covers learning, estimation, planning, and control for robots and autonomous systems. I combine technical direction and team leadership with hands-on algorithm development.

Planning and control

Planning a feasible path is only part of an autonomy problem; the system must also track it safely as conditions change. My work has included model predictive control, particle-filter-based motion planning, reachability, and invariant-set methods for autonomous vehicles, articulated vehicles, quadrotors, and mobile robots. In industrial settings, I have also led work on multi-robot routing, task allocation, and resource allocation.

Localization and estimation

Reliable autonomy depends on knowing where a system is and how it is moving. I have developed Bayesian filtering and sensor-fusion methods for vehicle state estimation, GNSS positioning, and joint vehicle localization and road mapping. This line of work includes particle filters, Kalman filters, factor-graph methods, and learning-based models, with attention to imperfect measurements and uncertain sensor noise.

Learning and adaptation

My work in learning includes online Bayesian learning of Gaussian-process state-space models, physically constrained neural system identification, and learning control policies from digital-twin simulations. The AIDA framework combines neuro-symbolic planning with residual learning for multi-task robot policies.

Mobile manipulation and robot autonomy

At Chewy, I lead autonomy algorithm work for mobile manipulation systems, from research through validation and engineering handoff. My work includes a robot-management system connecting mapping, scene understanding, task allocation, and routing, validated on physical robots. Earlier work at WASR and Symbotic addressed routing, control, and coordination in fleets of autonomous robots.

Research highlights

Full publication list
Experimental double lane-change plots on snow: a black dashed reference path, red dotted fixed-model response, and blue solid friction-adaptive response, with tracking and vehicle-state traces below.
On a snow-covered track, the friction-adaptive controller (blue) followed the reference path (black dashed) more closely through a double lane change than the controller using an asphalt tire model (red dotted).From Trajectory Tracking for Autonomous Vehicles on Varying Road Surfaces by Friction-Adaptive Nonlinear Model Predictive Control, Vehicle System Dynamics (2020). (opens in a new tab)
Three aligned timelines show when sensor measurements were timestamped, when a delayed measurement arrived, and which state estimates must be updated.
A delayed measurement arrives after newer observations. The estimator uses its earlier timestamp to update the current state, rather than treating arrival order as measurement order.From Rao-Blackwellized Particle Filters with Out-of-Sequence Measurement Processing, IEEE Transactions on Signal Processing (2014). (opens in a new tab)
A planned quadrotor path winds around red three-dimensional obstacles, beside photos of the indoor obstacle course and the Crazyflie quadrotor used in the experiment.
A quadrotor navigates through a course of obstacles using invariant-set motion planning. The adjacent photos show the indoor experiment and the Crazyflie vehicle.From Invariant Set Planning for Quadrotors: Design, Analysis, Experiments, IEEE Transactions on Control Systems Technology (2025). (opens in a new tab)

Research area

Mobile manipulation and robot autonomy

Methods that bring sensing, planning, and control together on physical robots.

Representative papers