Feelium: A Touchable Blimp Body for Aerial Telepresence

Technical Summary: Feelium – A Touchable Blimp Body for Aerial Telepresence

Problem Statement
Existing aerial telepresence systems, particularly drones, face significant limitations in facilitating natural social interaction. While research has shown that wrapping quadcopters in safe cages increases touch interactions from 1.4% to 38.9%, current systems fail to provide haptic feedback to the remote operator. Furthermore, traditional drones are constrained by loud rotors, safety risks, and transient, fingertip-scale contact sensing that relies on external tracking systems. Conversely, while blimps have been explored for decades as aerial carriers, screens, and companions (e.g., PRoP, LightBee), their envelopes have historically functioned as output-only devices. Although ground-based touchable inflatables exist, none have successfully connected a touched surface to a remote person, leaving a gap in creating a shared physical interaction channel for mediated closeness.

Methodology
The authors present Feelium, a system integrating a flying-display blimp platform, first-person VR inhabitation, and an onboard touch-sensing envelope.

• Hardware Platform: The physical body is a 64 cm indoor helium blimp weighing 323.7 g, with a buoyant lift of ~360 g. It utilizes four motors under closed-loop control for stable, quiet hovering. The envelope is largely blackened to retain helium and reject ambient light, except for a light-scattering front face acting as a rear-projection screen. This screen is illuminated by a 24.9 g laser projector (Ultimems HD309) via a 180° fisheye lens.
• Remote Inhabitation: A remote user wears an Apple Vision Pro, inhabiting the blimp in first-person. A front-facing fisheye camera serves as the "eyes," maintaining a wide field of view to keep a partner's reaching hand in frame until contact occurs. The user's face or avatar is projected onto the blimp's skin. Control commands (forward, backward, up, down, turn left, turn right) and media streams travel over HTTP.
• Touch Sensing Architecture: The system employs a dual-camera approach to sense contact on the inflatable skin:
  1. RGB Camera (Primary Decision Maker): A rear-mounted Raspberry Pi AI Camera (Sony IMX500) observes the membrane from the inside against the backlight. It uses a shadow pipeline that maintains a slowly adapting reference image of the untouched membrane. It computes per-pixel relative shadow depth (fraction of backlight lost) to ensure independence from uneven projector brightness.
  2. Thermal Camera (Presence Detection): A 32×24 MLX90640-D55 thermal camera senses heat. It detects warm bodies (hands/partners) approaching or hovering near the membrane by differencing frames against an ambient baseline but does not make touch decisions, as conducted heat through the membrane is too weak and slow.
• Touch Logic: The system distinguishes between "hover" and "touch" using two gates:
  • Instant Depth Gate: Detects firm presses immediately.
  • Sustained-Anchoring Gate: Computes the rolling minimum of shadow depth over the past second. This differentiates a pinned fingertip (sustained shadow) from a hovering hand (wobbling, thinning shadow).
  • Event Segmentation: A tracker segments decisions into discrete events (tap, hold, swipe) based on contact duration and fingertip path length.
  • Rendering: Detected touches are rendered as located markers in the VR user's view, and the projector responds at the specific touched spot.

Key Contributions
The paper claims three primary contributions:

1. The Feelium System: The first integration of a flying-display blimp, first-person VR inhabitation, and an onboard touch-sensing envelope for aerial telepresence.
2. Interaction Vocabulary: A defined set of two-party touch interactions between a co-located partner and the remote inhabitant, including greeting pats, palm-to-palm merging, cheek pecks, finger-drawing games, and lean-in hugs.
3. Preliminary Technical Validation: Empirical evidence demonstrating the feasibility of touch sensing on an inflatable envelope using the proposed shadow and thermal sensing pipeline.

Results
The system was validated using 228 labeled trials across three sessions, including 138 prompted touches (taps, holds, swipes) and 90 negative samples (hovers, disturbances, empty scenes), plus continuous clips.

• Detection Sensitivity: The shadow detector registered activity in 171 of 198 trials where a hand was present and never fired in empty scenes.
• Hover Rejection: Simple thresholding of the darkened area achieved 95% recall but only 77% precision due to false positives from hovering hands. The sustained-anchoring gate significantly improved performance, achieving an AUC of 0.99 in separating contact from hover on continuous clips. A threshold fitted on the first 60% of data held 0.90 balanced accuracy on the unseen remainder.
• Localization: For 69 detected taps and holds, the landing points clustered tightly around the prompted targets, with a median RMS spread of 9 pixels in the detector's 640×480 resolution.
• Performance: The processing pipeline runs onboard at approximately 15 fps.
• Limitations: Current results are limited to dim-room lighting and a single rectangular "sensitive skin patch" of the envelope.

Significance and Claims
The authors position Feelium as the first system to explore envelope touch in an aerial telepresence context. They argue that touch is a critical channel for deepening mediated closeness, moving beyond information exchange to "presence" as the payload. The paper claims the system successfully provides a physical interaction channel where the skin acts as a shared surface between remote and co-located partners. The work is presented as a preliminary technical validation that motivates further dyadic evaluation of social presence, whole-envelope sensing, haptic feedback to the wearer, and gesture-piloted flight, aiming to give remote presence a body that "invites touch" and brings separated partners within arm's reach.