Pooya Abolghasemi

About

Pooya Abolghasemi is a robotics and machine learning researcher whose work sits at the intersection of assistive robotics, learning from demonstration, and deep neural network-based manipulation. His research primarily focuses on enabling robots to assist disabled and elderly individuals by learning complex manipulation tasks directly from user demonstrations, removing the burden of technical programming from end users. Among his most significant contributions is his pioneering use of Long Short-Term Memory (LSTM) networks and Mixture Density Networks (MDN) to transfer manipulation skills learned in virtual environments to real-world robotic systems — a challenging domain adaptation problem addressed across several influential publications accumulating over 80 citations. His 2018 work on multi-task vision-based manipulation demonstrated that even low-cost robotic arms could master diverse picking, placing, and non-prehensile tasks through end-to-end learning from raw image input. Abolghasemi also advanced robustness in deep visuomotor policies through his attention-based approach, helping robots maintain task performance despite physical disturbances — a critical requirement for real-world deployment. His work on wheelchair-mounted robotic arms further demonstrates a commitment to practical assistive applications. Collectively, his research has meaningfully advanced the accessibility and reliability of learning-based robotic manipulation systems.

Research Focus

Key Achievements

Top Papers

1. 1
   From Virtual Demonstration to Real-World Manipulation Using LSTM and MDN

   37 citations · 2018
2. 2
   Learning real manipulation tasks from virtual demonstrations using LSTM

   28 citations · 2016
3. 3
   Pay Attention! - Robustifying a Deep Visuomotor Policy Through Task-Focused Visual Attention

   26 citations · 2019
4. 4
   Vision-Based Multi-Task Manipulation for Inexpensive Robots Using End-to-End Learning from Demonstration

   22 citations · 2018
5. 5
   From virtual demonstration to real-world manipulation using LSTM and MDN

   18 citations · 2016
6. 6
   Learning Manipulation Trajectories Using Recurrent Neural Networks

   15 citations · 2016
7. 7
   A Real-Time Technique for Positioning a Wheelchair-Mounted Robotic Arm for Household Manipulation Tasks

   10 citations · 2016
8. 8
   Trajectory Adaptation of Robot Arms for Head-Pose Dependent Assistive Tasks.

   5 citations · 2016
9. 9
   Real-time placement of a wheelchair-mounted robotic arm

   4 citations · 2016
10. 10
    Pay attention! - Robustifying a Deep Visuomotor Policy through Task-Focused Attention

    2 citations · 2018

Key Collaborators

Contact & Links