Keep your robots out of jail !

We're building Socially Situated Intelligence for Robots in Public Spaces.

Foundation models can reason about the world. They struggle to operate in it — reliably, at fleet scale, in spaces designed for humans. We investigate the points of interaction and the friction where robots break down, in order to build the runtime infrastructure that closes that gap.

For operators & OEMs
Building robots? Talk to us.

Humanoid OEMs, sidewalk fleet operators, service-robotics platforms. We work with a small number of partners before broader release. Tell us what you're operating today and where the failure modes hurt.

Contact us
StatusIn stealth
Founded2026 · San Francisco
ForRobots in public spaces
01 · The delivery case study
Commercial delivery case study: 16 social interactions a robot cannot handle Three packages on three floors across two buildings. The curve at top compares social-interaction load for a robot to a human baseline. The grid below catalogs each beat with its research-grounded capability and failure mode. Current robots can navigate geometry, but lack socially situated intelligence. In a typical commercial delivery there are at least 16 points along the workflow that require contextual intelligence. 3 2 3 16 ~15min PACKAGES BUILDINGS FLOORS SOCIAL INTERACTIONS HUMAN BASELINE EASY HARD HUMAN BASELINE ROBOT 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 UNLOAD DROP 1 DONE DROP 2 DONE DROP 3 Each numbered point is a moment requiring social interaction intelligence. The deeper the trough, the harder for a robot. EVERY BEAT, INDEXED DROP 1 · LAW FIRM, FLOOR 4 DROP 2 · BLDG B, GROUND DROP 3 · BLDG B, UPPER 01 Unload & ingress Egress from van, cross threshold AFFORDANCE PERCEPTION 02 Recognize reception Which person is staff vs visitor? OPEN-WORLD RECOGNITION 03 Interpret directions "That elevator, fourth floor, on the left." DEICTIC REFERENCE 04 Operate elevator Call, hold, select floor, share with riders AFFORDANCE + TURN-TAKING 05 Wayfind to suite Match label to door sign across hallway SYMBOL GROUNDING 06 Find suite empty No recipient. Now what? GOAL-STATE REVISION 07 Place & document Right spot, photo, geotag, log it PRAGMATIC INFERENCE 08 Exit through unknown door Different exit than the entrance DISTRIBUTIONAL SHIFT 09 Cross courtyard Outdoor route, no GPS detail SOCIAL NAVIGATION 10 Pace past pedestrians Match human pace, anticipate paths PROXEMICS & PACE-MATCHING 11 Enter Building B New lobby, new layout, new signage SCENE GENERALIZATION 12 Find ground reception Identify the desk, approach socially JOINT ATTENTION 13 Handoff & confirm Drop, sign, scan, photo MULTIMODAL GROUNDING 14 Operate elevator (again) Different building, different panel AFFORDANCE TRANSFER 15 Wayfind upper floor Third floor, third suite, third map SPATIAL MEMORY 16 Final handoff Place where? Hand to whom? When done? IMPLICIT INTERACTIONS SOURCES Gibson, J.J. (1979). Clark & Brennan (1991). Harnad (1990). Tomasello (1995). Hall, E.T. (1966). Ju, W. (2015).
Delivery is the case study. The pattern repeats wherever robots enter public space.
Last-mile delivery Hospitality Eldercare Security & patrol
02 · What we do

Operations is the product.

We are not building another pilot, another demo, another controlled-environment showcase. We are building the runtime that turns a robot platform into a fleet that operates — reliably, at scale, because it understands social context.

What we deliver

A perception runtime that generalizes across platforms, embodiments, and locations.

One system. Any robot. Any model behind it. Our runtime is designed to sit above π0, GR00T, Gemini Robotics, OpenVLA, or your in-house policy. Drop it in. Ship.

What we charge for

Per-robot, per-month. Aligned to your fleet, not your headcount.

You pay for the robots in service. We get paid when your fleet is operational. The incentives line up by design.

03 · Team

A decade of human-robot interaction in the wild.

Our team has studied human-robot interaction across the full spectrum of robotics — from autonomous vehicles at every level of autonomy, to robot trash cans on New York City streets, to commercial delivery operations at fleet scale. Public spaces. Real interaction. Real failure modes. This is the experience we are bringing to the problem.

Founder
WJ

Wendy Ju

CTO · Co-Founder
  • Professor at Cornell Tech; previously Executive Director of Interaction Design Research at Stanford CDR. PhD Stanford, MS MIT Media Lab. 10,000+ Google Scholar citations.
  • Wrote The Design of Implicit Interactions (2015) — the foundational text on how machines and people negotiate shared space without explicit instruction. We are productizing that work now.
  • Creator of the Robotability Score (CHI 2025) ranking how welcoming each NYC street is to delivery robots, and the BAD Dataset (with Accenture Labs) for detecting bystander reactions to robot failure in the wild.
Founder
NG

Nikhil Gowda

CEO · Co-Founder
  • Spent nearly five years as a Senior UX Researcher at Amazon Last Mile — daily field exposure to the largest delivery operation in the world, watching where UX actually breaks.
  • Decade in human-machine interaction research and field operations across delivery, automotive, and aerospace.
  • Holds 8 US patents; co-author and reviewer at Automotive UI, CHI, and HRI, with two papers in Nature Scientific Reports from human factors work at NASA Ames and the Center for Design Research at Stanford.
  • Co-author of RRADS (AutoUI 2015) and Nudge: Haptic Pre-Cueing for Automotive Intent with Wendy — ten years of collaboration on how machines communicate intent to humans in motion. It's what that decade of work was always pointing at.
Where our work has appeared
Cornell Tech Stanford CDR MIT Media Lab Toyota Research Institute NASA Ames CHI · HRI · THRI CNN · NPR · USA Today

Robots in the Wild is an infrastructure company. We make robotic fleets actually work — not in a demo, but on the metrics that decide whether the fleet stays funded.

The fleets we work with are trying to fill a labor gap and finding that they can't, because the operational inefficiencies between tasks compound faster than the unit economics improve. Robots are trained to perform tasks. The challenges are between them.

We are roboticists and social scientists working on the gap that doesn't show up in benchmarks — the one that has kept robots out of public space for forty years.

Come join us in building the most advanced socially situated intelligence in the world.

04 · Contact us

Tell us what you're operating.

We work with a small number of partners before broader release. Tell us about your fleet, your platform, and the failure modes that hurt — we'll get back within a few business days.

No mailing list. No automated follow-up. Just a real reply.