Show HN: OK-Robot: open, modular home robot framework for pick-and-drop anywhere
An open-source home robot framework called OK-Robot showcases how a relatively simple mobile manipulator can be instructed in natural language to find, pick up, and move everyday objects without per-home training. Commenters probe its technical limits—object recognition, failure on hidden items, slow and sometimes clumsy motion, reliance on relatively low-cost sensors—and note that while the software is open, the hardware platform remains expensive due to low production volumes. Many see the most compelling applications in assisting elderly or disabled people and offloading domestic chores, even as they highlight unsolved challenges like cluttered, obstacle-filled homes and broader economic questions about who ultimately benefits from automation.
Technical approach & capabilities
- System uses language-driven segmentation (Lang-SAM) plus CLIP embeddings to identify objects without predefining classes.
- Environment is mapped via RGB‑D cameras (e.g., iPhone with ARKit), back-projecting depth into a 3D voxel grid that stores both geometry and semantic embeddings.
- Framework currently focuses on two zero‑shot skills: pick and drop in unseen homes, contrasting with related work that supports more skills but needs per‑home demonstrations.
Limitations & open problems
- Cannot reliably:
- Detect infeasible requests (too heavy/large objects).
- Handle hidden or heavily occluded items or multi-step reasoning like “object is in a drawer in the kitchen.”
- Vision stack struggles when depth cameras give bad readings (e.g., transparent/sem-transparent objects).
- Motions look “fidgety” partly because the robot frequently re‑zeros its joints for accuracy; this is a tradeoff chosen with cheaper hardware and less sensing. Some argue encoders would reduce this at moderate extra cost.
- Robot is slow compared with humans; some see this as inherent safety/precision tradeoff, others simply question why general-purpose robots are so slow.
Cost, hardware, and economics
- Software is open source; hardware (e.g., Hello Robot platform) is proprietary and expensive (~$25k).
- Commenters note: low volumes, specialized parts, labor, calibration, and testing drive costs; patents are seen as less central than scale.
- Comparison to other industrial/specialty hardware where low-volume parts cost far more than similar mass-market parts.
Use cases, value, and social impact
- Enthusiasm for domestic help: laundry, dishes, cooking, cleaning, childcare-adjacent tasks. Some would pay far more than $25k if those were solved.
- Strong interest in benefits for elderly, disabled people, or overburdened caregivers; compared to the “curb cut effect” where accessibility improves life for everyone.
- Concern that such robots may primarily boost corporate profits and displace low-wage workers (e.g., hotel staff) without adequate social safety nets, sparking discussion of shorter workweeks, UBI, and dignity separate from labor.
Home environments & mobility
- Skepticism about wheeled robots in cluttered, real homes with cables, toys, pets, thresholds, and doors; some argue legged robots are ultimately needed.
- Others counter that “good-enough” wheeled systems plus minor behavior changes (clearing floors, simple guards for cords) can serve large markets now, with harder cases addressed later.