{ "schema": "pointcast.future-book-companion/v1", "id": "future-book-001-companion-05", "companionNumber": 5, "fieldReportNumber": 1, "blockId": "0579", "title": "The First Computer That Can Waddle", "subtitle": "Microduck, physical AI, and programming behavior with agents", "route": "/digital-pets/microduck", "jsonRoute": "/digital-pets/microduck.json", "publishedAt": "2026-08-31T16:10:00-07:00", "readingTime": "14 min", "status": "public-spec review of a pre-release product", "description": "A full review of Microduck, its open software stack, how to program learned behavior with AI agents, and the consumer arc from apps to abilities.", "url": "https://pointcast.xyz/digital-pets/microduck", "jsonUrl": "https://pointcast.xyz/digital-pets/microduck.json", "futureBook": { "title": "The Animal After the Internet", "url": "https://pointcast.xyz/digital-pets", "jsonUrl": "https://pointcast.xyz/digital-pets.json", "relationship": "Companion 05 and Field Report 01" }, "review": { "thesis": "The toy is the hook. The real product is a body for agents.", "verdict": "A development platform disguised as a character.", "preorderIf": "You want to train, modify, and break things in simulation first.", "waitIf": "You expect a polished household assistant that understands open-ended requests.", "ratings": { "approachability": 9, "softwareOpenness": 8.5, "outOfBoxAutonomy": 4, "builderValue": 8.5 } }, "disclosure": "Microduck had not shipped when this review was published. This is a review of launch hardware, public code, and the documented developer path—not a hands-on durability or battery test. Several specifications remained provisional.", "specifications": [ { "label": "Body", "value": "25 × 14 cm", "note": "Under 800 g; articulated legs, head, neck, and grasping beak." }, { "label": "Motion", "value": "15 DoF", "note": "Neural policies run through a 50 Hz onboard control loop." }, { "label": "Compute", "value": "RK3566", "note": "Rockchip processor with AI accelerator, 1 GB RAM, and 32 GB storage." }, { "label": "Vision", "value": "Camera + ToF", "note": "Front camera and compact 8 × 8 time-of-flight depth matrix." }, { "label": "Orientation", "value": "2 IMUs", "note": "One in the body and one in the head." }, { "label": "Interaction", "value": "Mic / speaker / NFC", "note": "Two NFC antennas and a generated voice unique to each unit." }, { "label": "Radios", "value": "Wi-Fi + Bluetooth", "note": "Gamepad control, local configuration, and networked development." }, { "label": "Power", "value": "Approximately 1 hour", "note": "Removable 2,600 mAh NP-F550 camera battery." } ], "architecture": { "principle": "Reflexes live in the duck; intent can live anywhere.", "loop": [ "human goal", "model plan", "agent contract", "50 Hz local policy", "camera and robot-state feedback" ], "boundary": "A frontier model should choose bounded skills, not improvise individual servo commands. Local control retains physical authority." }, "programmingPath": [ { "step": 1, "name": "Operate and observe", "outcome": "Learn the body before changing the brain." }, { "step": 2, "name": "Train a behavior", "outcome": "Write a reward rather than a choreography." }, { "step": 3, "name": "Program with AI", "outcome": "Describe the outcome and let an agent build and evaluate the loop." } ], "behaviorLab": [ { "id": "scout", "label": "Room Scout", "goal": "Find the red ball, approach it, and stop one body-length away.", "verbs": [ "gaze", "get_frame", "search_scan", "walk_to", "stop" ], "build": "Vision target plus bounded approach controller.", "reward": "Target centered; distance reduced; collision and unstable pitch penalized.", "gate": "20 randomized rooms; 95% clean stops; zero stair-edge entries." }, { "id": "bow", "label": "NFC Greeter", "goal": "When a familiar NFC tag taps the beak, bow once and quack softly.", "verbs": [ "nfc_read", "stand", "gaze", "quack" ], "build": "Event trigger plus head and neck motion policy.", "reward": "Smooth pose match; foot slip, joint speed, and repeat triggers penalized.", "gate": "500 simulated taps; head pitch capped at 25 degrees; 10-second cooldown." }, { "id": "flock", "label": "Flock Mode", "goal": "Two ducks find the ball together; only the better-positioned duck kicks.", "verbs": [ "search_scan", "broadcast", "walk_to", "yield", "kick" ], "build": "Shared task state plus local skill arbitration.", "reward": "Single clean kick and role clarity; duplicate motion and contact penalized.", "gate": "No double-kicks; radio-dropout recovery; local stop always wins." } ], "agentForwardProjects": [ "natural-language skill builder", "embodied memory", "reward-design partner", "flock orchestration", "skill provenance" ], "modelHorizon": [ { "layer": "Microduck ONNX policies", "role": "Balance, walk, recover, kick, and grasp at 50 Hz on-device.", "status": "shipping stack" }, { "layer": "General cloud or local language model", "role": "Interpret goals, select bounded skills, and write ordinary control software.", "status": "available now" }, { "layer": "Claude Mythos 5", "role": "Advanced coding and research capability; a signal of how powerful software agents are becoming.", "status": "restricted access" }, { "layer": "OpenAI Astra", "role": "Reported by OpenAI to show major gains in agentic coding and cybersecurity.", "status": "upcoming; not released" } ], "consumerArc": [ { "horizon": "now", "phase": "Playable body", "promise": "Buy the character and control it immediately." }, { "horizon": "1–2 years", "phase": "AI-authored skills", "promise": "Describe behavior and approve simulation evidence." }, { "horizon": "2–4 years", "phase": "Ability market", "promise": "Install skills with provenance and hardware envelopes." }, { "horizon": "4–7 years", "phase": "Household fleet", "promise": "Program the room through values, permissions, and limits." } ], "media": { "hero": "https://pointcast.xyz/images/digital-pets/microduck/morning.webp", "watching": "https://pointcast.xyz/images/digital-pets/microduck/watching.webp", "flock": "https://pointcast.xyz/images/digital-pets/microduck/kickabout.webp", "social": "https://pointcast.xyz/images/digital-pets/microduck/social-card.jpg", "photoCredit": "Pollen Robotics press kit", "socialCredit": "PointCast editorial illustration; not product photography" }, "sourceSatellite": { "url": "https://mhoydich.github.io/pointcast-microduck/", "repository": "https://github.com/mhoydich/pointcast-microduck", "relationship": "The standalone first edition remains available; this PointCast route is the canonical Future Book integration." }, "creators": [ { "name": "Michael Hoydich", "role": "Direction and editorial brief" }, { "name": "OpenAI Codex", "role": "Research, writing, design, and implementation collaborator" } ], "sources": [ { "label": "Pollen Robotics — Microduck product page", "url": "https://pollen-robotics.com/microduck/" }, { "label": "Pollen Robotics — Microduck press kit", "url": "https://pollen-robotics.com/microduck/press-kit/" }, { "label": "Pollen Robotics — Microduck runtime", "url": "https://github.com/pollen-robotics/microduck" }, { "label": "Pollen Robotics — Microduck RL", "url": "https://github.com/pollen-robotics/microduck_rl" }, { "label": "quackd — independent language-model planner", "url": "https://github.com/rokbenko/quackd" }, { "label": "Anthropic — Claude Mythos", "url": "https://www.anthropic.com/claude/mythos" }, { "label": "OpenAI — status and evaluation notes for Astra", "url": "https://openai.com/index/responding-next-frontier-critical-cyber-capabilities/" } ] }