Inspiration
Social connection should not require tolerating sensory overload or guessing unwritten social rules. We built Common Ground for autistic adults who want meaningful teamwork, strategic challenge, and a familiar group to return to. Calm presentation does not mean an easy or passive game: the difficulty comes from planning and coordinating with another person.
What it does
Teams of up to six build a persistent shared world. Two teammates enter each mission, choose responsibilities, plan robot routes, hold pressure plates to open gates, retrieve components, and return home before their rounds run out. Neither player can complete the cooperative mission alone. Successful missions restore an area of the world and credit the people who contributed.
Communication changes what happens: "Can you hold plate A?" creates an actionable request; a teammate can agree or decline. A player must wait until their partner crosses before leaving the plate. Map pings, short strategy signals, and optional free-text chat let teammates choose how to coordinate. Untimed turns, reduced motion, muted colours, and minimal text reduce unnecessary pressure while preserving strategic consequences. Either player can end an attempt if a partner becomes unavailable.
Two playable missions, Greenhouse and the more complex Power Station, demonstrate different maps and multiple plate dependencies. Team discovery is opt-in. Persistent team history makes each paired mission part of a larger group effort.
Why this matters
The game creates opportunities to practise requesting help, acknowledging commitments, clarifying a plan, and coordinating a handover with a real person. A stable team and shared record give players a reason to return to the same people. These are our intended pathways toward everyday communication and lasting bonds, not measured outcomes of this prototype.
Research on autistic adults' sensory experiences, gaming, and relationships informed the design. Evidence from other game-based social interventions is encouraging but does not establish that Common Ground improves real-world communication, and much of that evidence concerns younger participants. Autistic people have diverse preferences: players can choose structured signals or flexible chat. Our research sources and evidence boundaries are documented in the team specification.
How we built it
React and TypeScript run in a Vinext/Vite application hosted on Cloudflare Workers, with Cloudflare D1 persistence. A deterministic server-side engine resolves paired routes and pressure-plate/gate dependencies step by step. Version checks and request identifiers protect shared state from conflicting and repeated mutations. Clients poll for updates. Team progress, mission outcomes, and contributors persist across sessions. No AI model judges players or generates their conversations during gameplay.
Challenges we ran into
A gate must respond to both robots at each movement step: crossing should fail if the teammate leaves its plate too early. We also fixed mission launch synchronization, durable teammate requests, and completion persistence across independent browsers. We tested a complete deployed eight-round Greenhouse mission and verified that the restored area and both contributors survived reload.
Accomplishments and lessons
We shipped a working multiplayer browser game with a persistent shared world. During development, mandatory multiple-choice conversations felt detached from gameplay. We replaced them with communication that has real consequences inside the mission. The final design combines low sensory load with genuine interdependence rather than adding a social quiz to a puzzle.
What's next
Co-design and repeated-session testing with autistic adults, more missions, and stronger community safeguards. We would evaluate enjoyment, comfort, connection, and participant-defined goals, and assess transfer to unrelated collaborative activities separately from improvement at the maze. We have not yet conducted a user study or established a community partnership.
AI-use disclosure
OpenAI Codex assisted research synthesis, concept iteration, authored content, implementation, verification, and presentation preparation. ChatGPT, Google Gemini, and Grok were consulted for idea feedback. The shared-world illustration used the built-in image-generation tool. We began from the OpenAI Sites Vinext starter and use open-source dependencies. The team selected the direction and is responsible for reviewing and explaining the result. No AI model runs inside the shipped gameplay loop. Details: AI-USE.md.
