Amir Sadeghian
Papers
12
Total Citations
693
H-Index
8
About
Amir Sadeghian is a researcher whose work sits at the intersection of social robot navigation, trajectory forecasting, and deep learning for autonomous systems. His research has made significant contributions to how machines perceive, predict, and respond to human behavior in dynamic environments — challenges central to self-driving vehicles and socially aware robotics. Sadeghian is perhaps best known for his pioneering work on multi-agent trajectory prediction. His Social-BiGAT framework (2019, 278 citations) combined Graph Attention Networks with Bicycle-GANs to model complex social interactions between pedestrians, while SoPhie introduced attentive GAN architectures that respect both physical and social constraints in path prediction. Together, these works helped define a new generation of generative approaches to forecasting human motion. Beyond prediction, Sadeghian has contributed substantially to robot traversability estimation through GONet, a semi-supervised deep learning method leveraging GANs to assess safe navigation from camera images. His JRDB dataset (117 citations) provided the community with a rich, egocentric multimodal benchmark for human perception research, complemented by the real-time 3D multi-object tracker JRMOT. Collectively, his body of work reflects a coherent vision: equipping autonomous robots with the perceptual and predictive intelligence needed to navigate safely and naturally alongside humans.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3Deep Visual MPC-Policy Learning for Navigation88 citations · 2019
- 4
- 5
- 6
- 7JRMOT: A Real-Time 3D Multi-Object Tracker and a New Large-Scale Dataset17 citations · 2020
- 8
- 9
- 10