Swaying
A gentle lean guides the screen back toward the user’s line of sight.
RO-MAN ROBOT DESIGN COMPETITION / 2026
TILT
A Plant-inspired Tracking and Intelligent Laptop Teammate for Implicit Posture Correction
- TEAM
- Tsz Pi Liu · Sijin Qu
Zhengji Liang · Guiyin Tian - LAB
- JC-STEM Lab of Smart Cyber-Physical Industrial Systems
- CONTEXT
- RO-MAN Robot Design Competition
2026
THE IDEA
Posture support that moves with you.
Long desk sessions often lead to unconscious slumping, lateral leaning, and neck strain. TILT counters these small shifts with an ambient movement language: it leans, rises, or turns the screen toward a more comfortable viewing position.
MECHANICAL SYSTEM
Designed from the inside out.
A compact three-axis mechanism gives the laptop platform pitch, roll, and yaw. The final architecture combines the motors, camera, electronics, and power system in a low-profile base.
MOTION ARCHITECTURE
Growing upward
The platform rises to help restore a neutral viewing height.
Turning toward the sun
The display turns aside to invite a restorative break.
FORM LANGUAGE
Inspired by the quiet intelligence of a lotus leaf.
The sculpted top plate brings softness to a technical object, while the blue base references water. Together, they make TILT feel less like a machine and more like a responsive desk companion.
INTERACTION
It notices.
Then it responds.
Camera-based posture sensing is paired with small, understandable movements—providing feedback without a notification, alarm, or command.
Lean to the side
TILT leans in the opposite direction, helping return the screen to the user’s line of sight.
Forward slump
TILT rises with the user so the display remains easy to see while posture is readjusted.
Long work hours
TILT turns aside to signal a break, inviting the user to rest before continuing.
INTELLIGENCE LAYER
Two stages.
One gentle response.
TILT begins with fast, interpretable posture estimation, then asks a VLM to validate sustained changes before it moves. This two-stage approach keeps the interaction responsive while reducing false corrections.
VERSION 1MediaPipe-based estimationFast, interpretable landmark tracking+
MediaPipe detects upper-body and facial landmarks, then compares them with the user's calibrated baseline.
Landmark displacementflean = (Shoulder-midx − Shoulder-midx0) / Shoulder-wide0
Forward cuefforward = w1ΔFace-size + w2ΔFacey + w3ΔChin-to-torso
Rule-based actuationθmotor = Clip(kgain(fdeviation-range − tthreshold), θmin, θmax)Clip keeps movement within a safe angle range.
VERSION 2VLM-assisted validationValidate sustained posture changes+
When MediaPipe flags a possible change, the calibrated baseline and detected posture frames are sent to a cloud VLM via API.
The VLM validates the MediaPipe-triggered change against the calibrated baseline before actuation.
A short temporal capture window helps prevent bad postures from being missed by MediaPipe.OUTCOME
Less interruption.
More comfort.
TILT explores an alternative to explicit ergonomic reminders: a calm, physical interface that supports healthier posture through implicit interaction.
View project facts & design process ↗If you are interested in this project or have suggestions to share, we would be grateful to hear from you at u3659013@connect.hku.hk.