Insights
What was built, what was measured, what it means
Everything the lab publishes, in one place: experiments, field notes, technical writing, and the series.
Experiments
Experiments and benchmarks
Measured results from the bench: the hardware, the workload, what we measured, what failed, what we changed, and the result. Every number here also lives on a project page.
- One driving decision on the Thor, measured against the benchNVIDIA's open Alpamayo 1 reasoning model answered in 0.89 s on an RTX 5090 and 4.38 s on a Jetson AGX Thor. The Thor spent less energy per decision.
- A policy that passes in simulation has not passed yetOur manipulation policy completed 35 of 36 simulated turns. That earned it time on the real arm, not a checkmark.
- The runtime configuration cost more power than the modelOn Jetson AGX Thor, the same intrusion detector drew 54.1 W or 24.3 W depending on one threading setting. The model never changed.
Field notes
Field notes
Short observations from the bench and from sites: what broke, what held, and what it means for the next deployment.
Nothing published in this section yet.
Technical writing
Technical writing
Methods, architecture, and explanations for people who have to build and run these systems.
Nothing published in this section yet.
Newsletter by Obinna Edeh
At the Edge with Obyno
Conversations, demonstrations, and plain-language explanations of Physical AI, robotics, and edge deployment, hosted by Obinna Edeh.
Nothing published in this section yet. Until the first episode, the conversation is on LinkedIn.