Skip to content
[ gm.]
ROBOT PERCEPTION FOR DECISION-MAKINGROCHESTER, NY · 43.16° N
Open to Ph.D. research internships

Hello, I’m

Gesem
Mejia.

Researching robot perception and planning, sensor fusion, and world models.

Robot perception for decision-making: turning RGB, depth, sonar and audio into representations robots can plan and act on.

Ph.D. student at RIT · KAS LabPreviously 4.5 years at Intel Labs

Explore the work
FIG. 001
Conceptual visualisation of robot planning: a committed trajectory and alternative candidate paths threading between clusters of occupied voxels
PERCEIVE / PREDICT / ACTCONCEPTUAL ILLUSTRATION
EXPERIENCE
9yrs
Putting perception on physical robotsUAV · UGV · AUV
DEPLOYMENT
3fabs
Real-time multimodal system deployed worldwideMARIE · Intel Labs
LATEST RESEARCH
0.3→20.1%
Recall in heavy darkness with RGB–sonar fusionUnder review · RGB–sonar fusion
PUBLICATIONS
6papers
Journal, IEEE conference and workshop workEAAI · CoDIT · NeurIPS WS · SIGIR WS
FROM SENSORS TO DECISIONSSCROLL TO EXPLORE
Gesem Mejia
Gesem Eliab Gudino MejiaRochester, NY

Curious by nature.
Hands-on by design.

Robot perception for decision-making: turning multimodal sensor data into representations of the world that robots can act on for navigation and planning.

Ph.D. student in Electrical and Computer Engineering at Rochester Institute of Technology and Graduate Research Assistant in KAS Lab, advised by Prof. Mohammad Javad Khojasteh. Current research focuses on perception that withstands sensor degradation on underwater robots, fusing camera and sonar so that an AUV keeps perceiving its surroundings as visibility drops. Broader interests include world models (JEPA), diffusion models for planning, and MPC driven by learned perception.

Nine years of experience deploying perception on physical robots: quad-rotors, field rovers and AUVs. Previously, four and a half years at Intel Labs, architecting a real-time multimodal perception system taken from research prototype to deployment in three semiconductor fabs worldwide.

From sensors
to decisions.

Robot perception underwater, on the ground,
in the air, and in the fab.

Published under G. Mejia, G. G. Mejia, and G. Gudino.

2026

Sound Advice: Vision-Language-Conditioned RGB-Sonar Fusion under Camera Degradation

G. Mejia, A. Acharya, K. Abaid, M. Chaickowsky, and M. J. Khojasteh

Conference submission

UNDER REVIEW

From building
to understanding.

Nine years of putting perception on physical robots and into deployment, from a quadrotor to semiconductor fabs to AUVs.

EDUCATION

MAY 2026 — PRESENT

Ph.D. · Electrical & Computer Engineering

Rochester Institute of Technology

Fully funded graduate research assistantship
2019 — 2021

M.Sc. · Mechatronics

Centro de Investigaciones en Óptica

Specialization in path planning for robotics
2015 — 2019

B.Eng. · Mechatronics

Universidad Tecnológica de León

Minor in Optics
2026 — PRESENT

KAS Lab · Rochester Institute of Technology

Graduate Research Assistant · Advisor: Prof. Mohammad Javad Khojasteh

Perception under sensor degradation for AUVs: an RGB–forward-looking-sonar fusion detector that keeps the robot seeing when the camera fails. Ongoing: MPC with a control-barrier-function safety filter for AUV navigation among moving obstacles.

OCT 2021 — APR 2026

Intel Labs · Intelligent Systems Research Lab

Software Research Engineer / Scientist

Architected MARIE, a real-time multimodal action-recognition system learned from expert demonstrations and deployed across three fabs. Built C++ ROS point-cloud perception and 6-DoF pose estimation for Kidspace.

2019 — 2021

Robotics & Perception Lab · CIO

M.Sc. Researcher

Deep RL for obstacle avoidance in 6-DoF pick-and-place and UAV navigation in unknown environments; a stereo-camera rover on NVIDIA Jetson for strawberry detection and ripeness estimation.

2017 — 2019

Robotics & Perception Lab · CIO

Undergraduate Research Assistant

Real-time face detection and person-following control, flight-tested on a physical quad-rotor UAV.

IN MY TOOLBOX

Perception
  • Multimodal sensor fusion (RGB, depth, sonar, audio)
  • Object detection & tracking
  • Point-cloud processing
  • 6-DoF pose estimation
  • Stereo / RGB-D
  • OpenCV
Learning, planning & control
  • PyTorch
  • TensorFlow
  • CUDA
  • Self-supervised learning
  • Deep RL
  • MPC
  • Control barrier functions
  • Path planning
Robotics & simulation
  • ROS / ROS 2
  • Gazebo
  • MuJoCo
  • PyBullet
  • RViz
  • Simulink
Sensors & hardware
  • NVIDIA Jetson
  • Stereo & depth cameras
  • Forward-looking sonar
  • LiDAR
Languages & tools
  • Python
  • C/C++
  • JavaScript
  • Docker
  • Kubernetes
  • CI/CD
  • FastAPI
  • Qt

Where the work is headed.

Each direction starts from the work above.

01Multimodal perception

SO FAR

RGB–sonar fusion that keeps an AUV detecting objects in darkness and turbidity; multimodal action recognition deployed in three fabs; stereo perception on a field UGV.

NEXT

Perception that estimates how far to trust each sensor, across RGB, depth, sonar and audio, and passes that confidence on to the planner.

02Safe planning and navigation

SO FAR

MPC with control-barrier-function safety filters for AUV navigation among moving obstacles; CBF-guided diffusion planning, at an exploratory stage.

NEXT

Decentralized multi-agent navigation, and planners that act directly on perceived, uncertain scenes.

03World models for planning

SO FAR

No published work yet; the diffusion planner is the nearest groundwork.

NEXT

Predictive latent representations of the world, grounded in multimodal perception, that a planner can act on.

Have a question, an idea, or a research opportunity?

Let’s connect.

gg2673@rit.edu

Seeking Ph.D. research internships in robot perception for navigation, planning, or world models.