Amir Rasouli
I am Senior Staff Engineer at Noah's Ark Laboratory, Canada leading Embodied AI Team specialized in robotic manipulation and planning. Formerly, I was leading the decision-making and reasoning autonomous driving team, working on behavior understanding, scene reasoning, prediction, and planning. The focus of my current work is on robustness and generalization of robotic systems in real-world. In particular, I focuse on developing robotic manipulation policies, data generation, benchmarking, and simulation.
I have completed my PhD in computer science and M.A.Sc in computer engineering under the supervision of Prof. John K. Tsostsos at York University in 2020 and 2015, respectively. I received my B.Eng. degree in Computer Systems Engineering and B.A. in Business Management at Royal Melbourne Institute of Technology in 2010.
Research
Data Generation and Benchmarking
Data generation and benchmarking of robotic manipulation policies, in particular vision-action-language (VLA) models, with emphasis on robustness to clutter. Select papers: [ICRA26, arxiv26, ICRA26]Robotic Object Manipulation
Robotic object manipulation, in particular in highly cluttered environments with high chance of collision and safety hazards. Select papers: [ICRA26, NeurIPS25, IROS25]Trajectory Prediction for Autonomous Driving
- Vehicle and pedestrian behavior prediction for autonomous driving with a focus on scene multimodal trajectory generation, scene understanding, causal reasoning, and robustness study.
Select papers: [ CVPR24, ICRA24, ICCV23 ]
Pedestrian Behavior Understanding
- Human behavior understanding, including psychological studies of pedestrians in traffic and prediction of their trajectories and actions in assistive driving settings.
Datasets: [PIE, JAAD] Select papers: [ICRA23, ICRA22, ICCV21, ICCV19, Trans-ITS19, BMVC19, IV17]
Road User Behavior Simulation
- Methods for vehicles and pedestrian behavior simulation using learning based methods and heuristic approaches parameterized based on behavioral studies.
Select papers: [IV23, NeurIPS22, IV22]
Benchmarking and Metrics
- Approaches for data subclassing based on observed behaviors, agents' characteristics, and other contextual parameters. These are accompanied by novel metrics for more effective measurement of prediction and perception models' performance.
Select papers: [ICRA24, ICRA24, NeurIPS22, ECCVW18]
Active Visual Search
- Take advantage of contextual information and detect areas of importance via attention mechanism to optimize object search process.
Dataset: [3DGEMS] Select papers: [Autonomous Robots20, CRV16, CRV14]
Saliency Prediction
- Study the ability of methods for predicting saliency in images.
Dataset: [P3 and O3] Paper: [BMVC19]