MEMO

MEMO, developed by Sunday Robotics (operating under sunday.ai), is a pioneering wheeled humanoid home robot unveiled in November 2025, designed specifically for domestic chores in busy households. Unlike traditional legged humanoids that struggle with balance and energy efficiency, MEMO features a stable wheeled base with passive stability—ensuring it won't fall even if power is cut—and a height-adjustable torso that lowers to ground level or extends arms up to 2.1 meters vertically and 0.8 meters horizontally, resting at 1.7 meters (170 cm) for most tasks. Weighing 77 kg (170 lbs), it prioritizes safety with compliant actuators that yield to human touch, a soft silicone outer shell free of sharp edges, and IP66/IP67 ingress protection on arms and hands. The dual arms each offer 7 degrees of freedom (DoF), paired with 4 DoF pincer grippers (4 fingers) precisely mirroring the company's Skill Capture Glove™, enabling dexterous manipulation of delicate items like wine glasses or utensils. At its core is the ACT-1 foundation model, a revolutionary robot AI trained on zero teleoperation or robot-generated data. Instead, ACT-1 leverages over 10 million episodes of human movements captured from more than 2,000 Skill Capture Gloves worn by 'Memory Developers' in diverse real-world US homes—from studio apartments to family houses with pets and clutter. The glove, equipped with force sensors, cameras, and joint potentiometers, records 'memories' of chores, which are transformed via 'Skill Transform' (90% fidelity) into robot-equivalent trajectories. This end-to-end model integrates long-horizon manipulation with map-conditioned navigation, generating actions from millimeter-precision grasps to meter-scale movement, conditioned on 3D environment maps for zero-shot generalization to unseen homes. ACT-1 enables emergent behaviors like switching grasp types or handling novel objects intuitively, as if a 'Master Chef' observing diverse kitchens. Deployment history began in December 2024 with basic one-armed shoe arrangement, evolving by October 2025 to complex demos: autonomously clearing dining tables (33 unique interactions across 21 objects, navigating 130+ ft to load dishwashers, dump scraps), folding six pairs of socks from clutter (delicate pinches, stretching), and pulling espresso shots (bimanual tamping, high-torque insertion). These 5x-accelerated videos showcase room-scale autonomy without human intervention. Sunday's 48-hour iteration cycle—design, review, train (fine-tuning with data recipes/augmentation), evaluate in new environments—drives monthly skill library expansions. No user data is required for training; privacy is paramount, with consent for beta feedback. MEMO's architecture emphasizes pragmatic engineering: onboard compute (likely GPU-accelerated, Linux-based stack) for low-latency glass-to-action, collision avoidance for dynamic obstacles, and self-charging capability in beta. Current prototype costs ~$20,000/unit (hand-built), targeting 50%+ reduction via scale for 2027-2028 launch. Beta program (late 2026, invite-only) deploys limited units to 'Founding Families' for refinement in hygiene, durability, and interaction. This data-centric approach breaks robotics' scaling deadlock, positioning MEMO as the first truly home-ready helper, adapting to chaos without setup.