Radar / AI robotics / Perceptron Mk1.5
Perceptron Mk1.5: perception model for robotics adds audio and explicit annotation outputs
Perceptron shipped Mk1.5 (September 25), successor to its Mk1 vision-language model for video and embodied reasoning: adds audio as an input modality, expands context from 33K to 37K tokens, and formalizes annotation outputs (points, boxes, polygons, clips) as an explicit opt-in format; $0.15/$1.50 per million tokens via API and OpenRouter.
Why it matters
A purpose-built embodied-reasoning model that returns machine-parseable geometry (boxes, polygons) without a second parsing stage; the audio addition makes it multimodal for real robot loops.
What you could build with it
A warehouse-robotics team could run Mk1.5 on fixed-camera feeds to detect package anomalies and output bounding boxes directly to the controller, skipping a separate detection-parsing stage.
Does it hold up?
Too early to judge: no benchmark scores published and no independent production writeups yet; the value case rests on vendor-described annotation outputs.
Built with Perceptron Mk1.5
- Perceptron Mk1.5 vs Gemma 4 12B: which vision model winsarticle · Independent comparison concluding Mk1.5 wins where machine-parseable geometry output is needed, while Gemma 4 12B wins on context size and openness.
Learn more
First spotted on article: source.
More AI robotics
FLUX 3 ActionBlack Forest Labs launched FLUX 3 Action on Sept 24: a 7B open-weight world-action robotics model that takes camera…robotics · JEV 0.72NVIDIA Isaac ROS 5.0: agentic AI workflows for roboticsIsaac ROS 5.0 moves to ROS 2 Lyrical/Ubuntu 24.04 and adds agent-ready skills - reusable workflows like…robotics · JEV 0.59Light-O1 (Light Origins): Embodied Foundation ModelGeneral-purpose embodied foundation model that learns a reusable human action prior from structured human actions…robotics · JEV 0.43Isaac ROS 5.0GPU-accelerated robotics stack goes agentic with reusable AI agents for setup and manipulation tasks.robotics · JEV 0.32JBR-001: open-source 3D-printable desktop companion robotArduino UNO Q-powered desktop robot with camera and on-device computer vision trained on synthetic data, distance…robotics · JEV 0.25RAPID: Robot Agentic Programming from DemonstrationsA system that generates and verifies reusable robot programs from a single visual demonstration, tested both in…robotics · JEV 0.24
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.