// hello, world. or whatever planet you're on.
ayush zenith
research @ tera ai
building models that help real robots do real shit in the real world. no hacks. no excuses.
status: ● online · 42.37°N 71.11°W · cambridge, ma
whoami
I'm a researcher at Tera AI. We build spatial foundation models that let anything that moves figure out where it is and where it's going from a plain camera feed: drones, planes, boats, cars on-road and off-road, and basically anything else we can strap a camera to. No GPS, no clean maps, no per-platform tuning. It runs onboard, in real time, out in the real world, in daylight, at night, and over terrain with nothing to look at.
Before Tera I was bouncing between labs:
- GRAIL with Prof. Lawson Wong: mobile manipulation on a Boston Dynamics Spot. Vision-language models for perception, mapping, and planning, so it could navigate a building, open doors, climb stairs, and pick up and place objects along the way.
- Air Force Research Laboratory with Dr. Jing Lin: generated a lot of realistic synthetic data, trained vision models on it, and built SDQM, a metric that tells you how good a synthetic dataset is without training a detector to convergence.
- Genentech: deep learning for drug discovery and development. Large-scale multimodal semantic search engines and fine-tuned transformers.
And before that I was at Northeastern studying ML / AI / CS.
ls ./research
google scholar ↗ semantic scholar ↗
-
2026
SDQM: Synthetic Data Quality Metric for Object Detection Dataset Evaluation
Journal of Electronic Imaging 35(3), 2026
-
2025
TARDIS STRIDE: A Spatio-Temporal Road Image Dataset and World Model for Autonomy
arXiv preprint, 2025
-
pat.
Spatially Consistent Geolocation Model
US Patent Application 19/339,112
tail -f tera.log
recent transmissions from the Tera blog:
- Your GPS-denied nav "demo" is likely still using GPS
- Navigating in the dark: GPS-denied visual navigation on LWIR
- Delivering autonomy in days instead of months via software
- Pushing the limits of visual navigation in ultra-sparse environments
- Universal navigation for every moving robot
- TARDIS STRIDE: towards rich real-world understanding
cat ./off-hours.txt
When I'm not staring at trajectories, I'm shooting on my Fujifilm (@yush.jpg), sparring after 10+ years of taekwondo and karate, occasionally rolling jiu-jitsu, keeping strongly held food opinions on Beli, losing at chess, and watching too much anime.
ping ayush
Working on robots, navigation, or anything spatial? Say hi.
- ayushp.zenith@gmail.com
ayush@tera-ai.com - scholar
- google scholar / semantic scholar
- github
- @ayushzenith
- in/ayushzenith
- x
- @AyushZenith
- @yush.jpg