Available for research collaborations

Building robots
that feel the world.

$

I'm Elijah — co-founder & CTO of Devol Robots, where we're teaching machines to understand contact, adapt in real time, and work alongside humans. Outside the lab, this site is where I share research, half-baked ideas, and notes from the frontier of embodied AI.

Explore research Say hello

01 / About

A roboticist who'd rather build than talk about it.

I trained as a roboticist at Oregon State University, where my graduate research focused on giving robot hands a sense of touch — combining deep reinforcement learning with human grasp demonstrations to make manipulation robust under real-world uncertainty.

That work led me to a single conviction: the next generation of robots will be defined by how well they sense and respond to physical contact — not by how fast they can run a pre-recorded trajectory. So I'm building that future at Devol Robots.

When I'm not in the lab, you'll find me reading papers, watching stand-up, or chasing some new corner of the world. I believe in growing by being uncomfortable, and in working with people who do the same.

Discipline

Embodied AI

Manipulation, RL, force control

Publications

IROS · ICRA

Peer-reviewed robotics venues

Role

Co-founder · CTO

Devol Robots Inc.

Based

Earth, 2026

Currently building hardware

02 / Now

What I'm working on.

The short version: making robots that can do real, contact-rich work in the wild — not just on rails in a factory cell.

→ Co-founder · CTO · Devol Robots

Robots that understand contact. Adapt in real time.

At Devol, we design force-control robotic systems with proprietary actuators and an embodied AI stack built around contact. Our platform tackles tasks that have stumped conventional automation for decades — plug insertion, hearing-aid assembly, snap-fits, DIN-rail work — by giving robots an intuition for force, alignment, and recovery.

I lead the technical side: hardware, controls, the learning stack, and how those pieces become a product. It's the kind of problem you can only crack by working from atoms to algorithms in the same building.

Visit Devol Robots

01Force-control actuators that feel and adapt — not stiff position controllers pretending to.
02Embodied AI models trained on real contact data, not just simulation.
03Assembly tasks that traditional automation can't touch: plug insertion, snap-fit, alignment.
04Building a platform — hardware + software — that ships to real customers.

03 / Research

Selected research & projects.

A sampling of past work, mostly in robotic manipulation, grasping under uncertainty, and benchmarks for hand design. The full archive is on the projects page.

Robot grasping under uncertainty

Master Thesis · RSS 2020

Robot grasping under uncertainty with deep RL

Combining human-adapted grasp strategies with Deep Deterministic Policy Gradient to teach robot hands to grasp varied objects from noisy in-hand poses.

↗ github RL · manipulation

Robotic manipulation benchmark

A pose-perturbation benchmark for grippers

A new benchmark for evaluating gripper hardware and grasp planning, with autonomous scene reset and binary-search-driven trial sequencing on a physical testbed.

↗ github hardware · evaluation

Near-contact grasping study

Near-contact grasping: humans vs. robots

Human studies on multi-robot hands revealed three primitive grasping strategies under uncertainty. I distilled those into PID controllers and presented the work at IROS.

↗ paper ↗ video ↗ github

Grasp metric from geometry

Grasp quality from geometric relationships

Instead of relying on noisy haptic contact data, we trained a grasp-quality classifier from the geometric relationship between robot hand and object surfaces across hundreds of trials.

↗ paper geometry · ML

Physical grasp testing infrastructure

Autonomous grasp-testing testbed

A reset-and-sense rig that automates robot grasp trials end-to-end, enabling massive real-world data collection and lowering the sim-to-real barrier.

Yale OpenHand build

Hardware · Undergrad

4-DOF Yale OpenHand Model O

Built an open-sourced underactuated 4-DOF gripper and wrote the control interface to actuate it. Where my robotics journey actually started.

See all projects

04 / Experience

The path so far.

A timeline of the roles and rooms I've worked in. Full version on the experience page.

2023 → Now

Co-founder & CTO

Devol Robots Inc.

Leading hardware, controls, and the embodied-AI stack behind our force-control assembly platform. From actuator design to deployed product.

2020 → 2023

Senior Robotics Engineer (AI & Software)

Roboligent Inc.

Led software, AI solutions, and robotics control across the platform — from learning-based perception to low-level controllers on the company's force-aware manipulators. The years that turned research instincts into shipping systems.

2018 → 2019

Graduate Researcher

Oregon State University · Robotic Manipulation

Master's thesis on grasping under uncertainty with deep RL, advised by Profs. Grimm, Fern, and Balasubramanian. Published at IROS & ICRA.

2017

Mechanical Engineer Intern

A.K.Ong Engineering · Singapore

HVAC system design, on-site supervision, and cross-team coordination. First taste of real-world engineering tradeoffs.

Full experience

05 / Writing

Notes & field reports.

A slow-burn blog about robotics, AI, and the messy reality of building physical machines. Subscribe via RSS or check back when curiosity strikes.

First post landing soon.

I'm prepping a series on what we've learned at Devol — what works, what surprised us, and where embodied AI is actually headed.