Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces the Drones2BodyLanguage dataset, which enables aerial robots to perceive human motion and intent by grounding avatars in real UAV footage with accurate positioning and orientation. Utilizing a lightweight geometric world model, the authors predict placement points for avatars representing ten communicative intents, achieving significant improvements in intent recognition across various architectures. The results demonstrate that training on this dataset enhances mean intent accuracy substantially, even in real-world scenarios, highlighting its potential for advancing socially intelligent aerial robotics.
Training aerial robots to understand human body language just got a major boost with a dataset that grounds avatars in real UAV footage, achieving unprecedented intent recognition accuracy.
Perceiving human motion and intent at long range is a prerequisite for socially intelligent aerial robots, yet the data to learn it barely exists. We introduce Drones2BodyLanguage, a dataset grounding human motion in real UAV footage: avatars manifesting ten communicative intents are placed into unmodified 4K drone scenes with metrically correct position, scale and orientation, maintained over hundreds of frames of camera motion. Enabling it is a lightweight geometric world model of the local scene - semantically selected anchors lifted to 3D through streaming monocular depth - in which a placement point is predicted as an affine anchor combination with provably rigid-invariant weights, and re-rendered under an SVD-fitted ground rotation. Across twelve architectures on scene- and motion-disjoint splits, training on placed data lifts mean intent accuracy by a wide margin for real, retargeted and generated motion alike, with gains confirmed on two in-the-wild scenes.