Senior Robotics Engineer at Neya Systems, working on robust software for off-road autonomy. I am the team lead and a core contributor to Neya's Virtual Integration and Simulation Environment (VISE), a high-fidelity Unreal Engine-based off-road autonomy simulator.
I was previously at RE2 / Sarcos Robotics as an Autonomy and Motion Planning Engineer. My work there spanned embedded software for manipulators, hierarchical state machines, motion planning and control for manipulation, and computer vision.
My research background is in robotics, applied mathematics, and mechanics. I most recently co-first-authored GraphEQA (published in the 2025 Conference on Robot Learning), a novel approach to embodied question answering using real-time 3D metric-semantic scene graphs and vision-language models. Before that, I worked on geometric methods for control and coordination of biologically inspired multi-robot systems in the Biorobotics Lab at Carnegie Mellon University.
Reinforcement Learning and Interstellar's TARS Robot in MuJoCo
TARS, from Christopher Nolan's Interstellar, is a rectangular-prismatic robot whose locomotion is governed by the rotation of rigid rectangular links about shared edges, a kinematic structure with few analogs in the robotics literature. This project was motivated by a pretty basic question I asked when I saw TARS locomote in the movie; "would it actually move like that?". My aim is to investigate what locomotion emerges for a TARS-inspired robot using model-free deep reinforcement learning.
I authored USD and MJCF models of a simplified TARS-like robot and trained locomotion policies using MuJoCo Playground, evaluating the effect of algorithm choice and reward parameterization on the structure and stability of the resulting gaits. This project is ongoing! See the GitHub page for details.
GraphEQA: Using 3D Semantic Scene Graphs for Real-time Embodied Question Answering (2024 - 2025)
Left panel: Robot equipped with GraphEQA, an approach that utilizes real-time 3D metric-semantic scene graphs (3DSGs) and task relevant images as multi-modal memory for grounding Vision-Language Models (VLMs) to perform EQA tasks in unseen environments. The robot uses GraphEQA to navigate a home while planning and mapping the environment in real-time; captured via an externally mounted camera.
Right panel: Metric-semantic 3D mesh and scene graph construction from Hydra. TSDF-based 2D occupancy map where white nodes represent explored areas, red nodes are obstacles, blue nodes are clustered frontiers. Green node shows the target location chosen by the planner.
Right panel inset: Video feed from the robot head camera.
-
GraphEQA: Using 3D Semantic Scene Graphs for Real-time Embodied Question Answering
-
The Geometric Structure of Externally Actuated Planar Locomoting Systems in Ambient Media
-
Stability and Control of Chaplygin Beanies Coupled to a Platform Through Nonholonomic Constraints
-
Locomotion of a multi-link non-holonomic snake robot with passive joints
-
Mechanics and Control of Coupled Interactions in Ambient Media