Event
TrajMamba: An RGB-D Human Trajectory Prediction Model
1 reports · 1 sources
Basheer Al-Tawil and co-authors propose TrajMamba, a lightweight bidirectional state-space model based on RGB-D cameras for short-term human trajectory prediction for mobile robots. The authors state that the model uses the same processing path for both training and deployment, aligning human pose extraction, depth back-projection, and smoothing. The authors state that TrajMamba has only 262K parameters, can run in real time on ordinary hardware, and has been deployed on a TIAGo robot equipped with an NVIDIA Jetson Orin.
Report timeline
- TrajMamba: A Real-Time RGB-D Human Trajectory Prediction Model for Mobile Robots
The authors propose TrajMamba, a lightweight bidirectional state space model for short-term human trajectory prediction using an RGB-D camera on mobile robots. Training and deployment share a single processing path to align human pose extraction, depth back projection, and smoothing.
arXiv cs.RO ·