Movement coaching
Use a camera to estimate arm positions and angles. Compare them with an exercise database and the person’s starting point, then give feedback on individual targets.
MOVEMENT COACHINGFABRICATION LAB / PROJECT PROPOSAL
A personal
training assistant.
Torto uses a mobile robot to support personalised exercise between physiotherapy sessions.
For people with limited mobility, individual guidance may be hard to access between appointments. We propose feedback without attaching wearable sensors to the body.
BUILT ONTurtleBot3 Burger/DESIGNED FORPersonalised exercise
FIG. 01 / TORTOProposed printed enclosure
FABRICATION LAB · INITIAL CONCEPTTurtleBot3 Burger · Camera · mmWave · Environmental sensorsPROJECT OVERVIEW
01 / THE CORE IDEA
Movement first. Context alongside it.Three functions.
One training session.
Breathing at rest
Explore contactless breathing measurements with mmWave radar during still rest periods between exercises.
BREATHING MONITORINGRoom conditions
Measure temperature and CO₂, and display alerts when agreed limits are exceeded—especially useful when uncomfortable conditions may go unnoticed.
ENVIRONMENTAL MONITORINGProposed capabilities to develop and validate. Personal targets reflect current abilities and are agreed with a physiotherapist.
02 / PROJECT SCOPE
Core prototype and optional extensionTwo configurations
THE CORE PROJECT
Torto
The core functions we plan to bring together.
- Arm-angle estimates and individual targets
- Reference exercises and progress across sessions
- Breathing observations during still rest
- Room temperature and CO₂ alerts
OPTIONAL EXTENSION
Torto Plus
Everything in Torto, plus a thermal camera.
Thermal camera
Explore skin-temperature changes as an additional source of information during training.
Intended to support a physiotherapist’s training plan. Thermal sensing is exploratory; skin temperature is not core body temperature.
03 / OUR PROOF OF CONCEPT
One seated exercise. One individual target.A seated arm exercise
Position the robot, then measure while stationary.
The robot moves to a suitable viewing position, stops, and observes one seated arm exercise. It estimates movement angles, compares them with a stored reference and the person’s baseline, and displays feedback.
STARTING POINT50°
AGREED TARGET55°
Review progress across sessions and export a summary for your doctor or physiotherapist. During a still rest period, display breathing and room measurements.
Illustrative goal agreed with a physiotherapist · not live measurements
ARM-RAISE ILLUSTRATIONREADY
EXAMPLE PROMPTLet’s raise one arm.
0 / 1SENSOR DISPLAY / INTERACTIVE EXAMPLE
Training session preview
Example reading
Example reading
Illustrative estimate · still rest
Baseline 50° · agreed target 55°
Fictional readings illustrate the proposed display. The arm value comes from the example exercise; breathing is observed during rest.
SESSION REVIEW / TORTO PLUS PREVIEW
Review progress. Take it to your next appointment.
Export a session summary and the underlying readings to discuss with your doctor or physiotherapist.
Thermal view
SYNTHETIC IMAGE20°C24°C28°C32°C36°C
Move over the image to inspect a simulated pixel.
Arm-raise progress
3 EXAMPLE SESSIONS| Session | Peak angle | From baseline |
|---|---|---|
| 01 | 50° | — |
| 02 | 51° | +1° |
| 03 | 52° | +2° |
Example history stays separate from the current sensor scenario. A measured change must exceed the system’s uncertainty before being interpreted as progress.
Your data, shared by you.
The CSV includes the three example sessions, the selected sensor snapshot, availability flags, demo thresholds and the synthetic thermal range. The printable summary includes this display. Downloading sends nothing to anyone.
DEMONSTRATION REPORT · FICTIONAL DATA · NO PATIENT RECORD
Adjust the demo alert thresholds
Upper thresholds for this fictional example only. Actual limits must be agreed for the person and setting; these are not health or safety recommendations.
Enter a temperature from 0 to 50°C and CO₂ from 300 to 10,000 ppm. These are input bounds for the demo.
No automatic breathing or exercise safety thresholds are set. A training target is not a safety limit. Changes apply only to this page until it is reloaded.
04 / PHYSICAL DESIGN
Three interpretations of the same printed shellEnclosure studies
01 / LiDAR stays exposedAll cover edges and brackets below the horizontal scan plane.
02 / Make it in partsRemovable panels, visible fasteners and shallow surface details.
03 / Give sensors spaceOpen camera view, ventilation and access to the electronics.
Selected concept: Turtle — brown panels with olive-green borders.
Proposed FDM covers, not validated CAD. Measure the actual robot and LiDAR, then verify scan clearance, wheel travel and camera view before printing. ROBOTIS LiDAR reference
05 / FROM IDEA TO PROTOTYPE
Prove the measurements before relying on them.Development plan
and sensor limitations.
01 / POSITION
Position the robot
Move to a suitable position and stop before observing the seated exercise.
02 / PERSONALISE
Establish a baseline
Store a reference exercise, an initial measurement and an agreed individual target.
03 / COMPARE
Compare measurements
Estimate arm angles, display feedback and record progress across sessions.
04 / REST
Observe rest periods
Observe breathing while still and display room conditions. Explore thermal sensing later.
SENSOR LIMITATIONS / VALIDATION PRIORITIES
Limits of contactless measurement
Breathing is inferred from motion.
- Body movement can mask breathing. Exercise, talking or robot vibration can interfere with the small chest movements used for estimation.
- Position matters. Distance and the view of the chest affect the signal. The Burger sits low: mounting height and radar tilt need testing without obstructing the LiDAR.
- The scene matters. Nearby moving people or objects can interfere. A general presence-detection radar is not automatically a validated breathing sensor.
Our approach: one seated person, a stopped robot, still rest periods and a tested measurement window. Suppress unreliable estimates.
Source: TI vital-signs demo limitationsSurface temperature, not core temperature.
- Only the visible surface is measured. Clothing hides the skin underneath; ordinary glass is not a suitable viewing window for typical long-wave thermal cameras.
- Readings depend on conditions. Surface emissivity, reflected heat, ambient conditions and calibration affect temperature estimates.
- Detail depends on the camera and distance. Small skin regions may occupy too few pixels for a useful reading. Sensitivity to small changes is not the same as absolute accuracy.
Our approach: explore exposed-skin temperature changes under consistent conditions. Keep thermal sensing optional; do not infer fever or overheating from a skin reading alone.
Source: FLIR measurement guidance · Surface visibilityWhat we must confirm before choosing the sensors
mmWave: breathing-capable processing, working distance, field of view, mounting orientation, time needed for an estimate and a way to flag poor signal quality. Test against a reference during rest and with deliberate motion interference.
Thermal: whether the module provides calibrated temperature data, resolution, field of view, focus distance, accuracy, warm-up requirements and calibration. Check whether the intended skin region is large enough in the image.
No universal range or accuracy is claimed here: specifications depend on the chosen module and processing. These manufacturer references describe their own devices and demos, not validated performance of our robot.
Camera angle accuracy, room-sensor placement and consent for session recording also remain prototype checks.
Support between physiotherapy sessions.Measure individual progress against an agreed training goal.
What we need to test: angle accuracy · camera placement · breathing reliability · useful feedback · sensor integration