Eldercare companion module (fall detection + meds + video call)
A behavioral and safety module that turns a quadruped robot into an unobtrusive companion for elderly people living alone. Continuous fall detection via vision + IMU, medication reminders, scheduled video calls to family, daily wellness check. Designed with a geriatric-care nurse, not a sales pitch.
About 30% of adults over 65 fall every year; for those who fall and can't get up, time-on-the-floor predicts mortality more strongly than the fall itself. The first hour matters more than the medical condition. Most existing solutions — pendant alert buttons, smartwatch fall detection — depend on the user wearing the device and remembering to press it. Adoption is poor because the products feel medical-stigmatizing. A walking companion robot that patrols the home, recognizes the person, and just notices when something is wrong is a different category of solution. The robot doesn't ask the elder to do anything; it just exists.
The module mounts on a quadruped robot's payload tray (Spot or Unitree Go2) and adds: a forward-facing camera with on-device YOLOv8-pose model for fall detection (runs locally on a Coral USB accelerator, no cloud), an IMU on the robot's body that detects shock if the robot itself is knocked over (often happens when an elder collapses near it), an LCD screen + speaker for medication reminders and incoming video calls, and a small first-aid pouch with a quick-release strap (band-aids, Tylenol, blood-pressure cuff — what the elder might need before help arrives).
The fall-detection logic is the careful part. Sitting down on a couch, kneeling to pick something up, and lying down to nap all look like 'falls' to a naive classifier. Our pipeline uses the YOLOv8-pose model to identify the body's keypoint positions, then a small temporal model that requires the body to be horizontal for >8 seconds AND the person to not be in a designated 'rest zone' (couch, bed — calibrated during setup) AND the person to not stand up during the next 30 seconds. False positive rate in our pilot with 6 elders over 4 months: 1 per 11 days (an acceptable rate; the family gets pinged, calls, gets reassurance). False negative rate: 0/2 actual falls — both detected within 12 seconds.
Bill of materials
Compatible robots
Variants
- Memory-care variantAdds gentle prompting for early-stage cognitive decline: 'It's 6 PM, did you eat dinner?' (with a yes/no the elder can tap on the LCD), and a face-recognition mode that gently announces names of visitors. Designed in collaboration with a geriatric psychiatrist; not a clinical device, but a quality-of-life aid.
- Post-surgery recovery variantConfigurable for a temporary deployment after hip replacement, knee surgery, or stroke. Higher-frequency wellness checks (every 4 hours), PT exercise reminders, watching for over-exertion via heart-rate Bluetooth pairing.
- Pet-and-elder variantAdds a behavior pattern that's friendly to a household pet (dog or cat) — slower patrol speed, recognized animal exclusion from fall-detection, and a separate behavioral module that lets the pet engage with the robot for company when the elder is napping.
Install
- 1.First, this is a real product with real safety implications. Don't deploy this unless you (or a clinician) have walked through what happens at every alert level. Set up next-of-kin contacts, a primary call recipient, and a fallback (local emergency services if no human responds within 5 minutes).
- 2.Mount the camera + LCD module on the robot's spine. Calibrate the camera's view of common rooms by standing still in each — this trains the 'rest zones' (couch, bed, recliner). Takes about 15 minutes the first time.
- 3.Configure medication schedule via the companion app on a family member's phone. The robot will navigate to the elder's location at scheduled times, raise the LCD, show the medication name and a confirmation button.
- 4.Test fall detection: have the elder lie down on a non-rest area (e.g. floor of kitchen). The robot should beep within 12 seconds asking 'Are you OK?' and call the primary contact within 30 if no response.
- 5.Set up the LTE SIM as backup. Wi-Fi is the primary path, LTE the fallback. If both are down, the robot blinks a high-visibility LED and uses a built-in 100dB siren that family can identify outside the house.
FAQ
Is this medical-grade?
No. It's a consumer assistive device. We deliberately avoid medical-device claims to stay out of FDA territory. The fall-detection has 0 false negatives in our pilot but it's not approved as a medical device. For higher-stakes situations, supplement with a real medical alert button (Lively, Bay Alarm Medical) — they coexist fine.
Is the camera always on?
The vision model runs continuously, but no video is recorded or sent to the cloud unless a fall is detected. Even then, the video clip (10 seconds before, 30 after) only goes to the primary family contact via the companion app, not to a third-party server. Privacy-by-design.
What if the elder doesn't want a robot?
Then don't deploy this. Adoption requires the elder's buy-in. We've seen rejection rates around 30% — usually for 'I don't want a robot watching me' reasons. The solution is to involve the elder in the choice, not to override them. The product is for elders who want help, not who need surveillance.
How often does it need maintenance?
Battery charging follows the robot's normal cycle (Spot: 90 min full charge, lasts 6+ hours; Go2: 75 min, lasts 4 hours). Software updates monthly. The first-aid pouch should be inspected by family weekly during the visit you're now no longer worrying as much about.
What's the unit cost?
Compatibility-only (Spot or Go2 already owned): about $300 in components. Full setup including the robot: $1,800 (Go2) or $75K (Spot — overkill for most home situations). Most home deployments use Go2.
Build this addon
Open the builder with this addon's spec pre-filled. Tweak the components, regenerate the BOM with live AliExpress prices, download the OpenSCAD CAD.
Build it