01
Ingest the robot's reality
RGB, depth, LiDAR, pose, IMU, odometry, and robot-state streams enter one timestamped spatial pipeline.
Robot world generation
SenseSight
Live 3D world models from the robot's own sensor stream.
See what the robot sees. Understand what it knows. Decide what it should do. SenseSight pivots the Human Sense thesis toward the world-generation layer: real-time spatial memory that a person can inspect before the robot acts on it.
- LIVE
- sensor-to-world loop
- 3DGS
- splat-ready architecture
- HITL
- reverse singularity review
- OSS
- open by default
The product
The world model is the interface
SenseSight turns raw robot streams into a spatial scene a person can enter: not a dashboard of disconnected telemetry, but the robot's working model of the place.
Live capture Clark Center Atrium
RGB 60 fps Depth aligned Pose tracking World streaming
02
Generate a living world
Frames become explorable geometry, trajectories, risk surfaces, semantic anchors, and splat-ready artifacts.
03
Keep the human in the loop
A person can inspect exactly what the robot sees before the system treats the world model as operational truth.
Realtime loop
See what the robot sees while the world is still changing
The promise is not a polished reconstruction after the mission. It is a live loop where a person can watch the model form, catch uncertainty, and make the machine better in the moment.
- 01
Robot streams
The robot publishes synchronized perception, pose, and health signals while it moves through the physical site.
- 02
World builds
SenseSight fuses each packet into a continuously updating 3D scene that can be replayed, queried, and audited.
- 03
Person sees
The operator sees the machine's evidence in spatial context instead of guessing from isolated camera frames.
- 04
System learns
Approved corrections, labels, and route judgments become durable memory for the next mission pass.
From pixels to place
RGB-D capture becomes navigable spatial memory.
SenseSight is built for the moment raw perception crosses into usable world state: point previews, splat refinement, semantic labels, and human corrections flowing back into the map.
Architecture
Open foundations for a serious robot-world platform
The baseline repo stays small, typed, and deployable. The package boundaries leave room for live viewers, robot adapters, storage, splat assets, and operator review without turning the site into the app.
Astro front end
Fast static marketing surface today, ready for richer product routes as the demo matures.
Typed spatial core
Shared TypeScript contracts keep observations, risks, decisions, and mission events consistent across apps.
Cloudflare edge
Pages deployment, preview branches, and room to add Workers, R2, D1, and realtime APIs behind the same domain.
Gaussian-splat path
Designed around point-cloud previews now and higher-fidelity splat assets as real robot reconstructions land.
Robot data adapters
Clear boundaries for OpenLORIS, live robot feeds, simulator missions, and future customer capture formats.
Human review trail
Reverse singularity is an architecture requirement: human judgment improves the world model instead of disappearing.
Open source
A public repo with production habits from the first commit
SenseSight starts as an open-source monorepo with CI, typed contracts, preview-friendly Cloudflare deployment, and a roadmap that can absorb real robotics data without a rewrite.
Reverse singularity
Robots generate worlds. People keep them honest.
SenseSight is the infrastructure layer for a human to inspect, correct, and trust a robot's evolving model of reality. The repo is structured so marketing, contracts, adapters, and future realtime surfaces can grow independently.
View the repositorySenseSight.live
Build the robot's world before asking it to act.
A real-time world-generation layer for robotics teams who believe the future is neither blind autonomy nor manual teleop. It is machine perception sharpened by human judgment.
Open sensesight.live