July 20 - Best of ICRA (Day One)

The Best of ICRA is a three-day virtual meetup series featuring researchers presenting their accepted papers from the 2026 International Conference on Robotics and Automation (ICRA).

Register for the Zoom to get access to all three days of the Best of ICRA.

Each session features a curated lineup of speakers sharing cutting-edge research across robotics, computer vision, and AI — straight from papers accepted at one of the field's top conferences.

Towards Versatile Opti-Acoustic Sensor Fusion and Volumetric Mapping for Safe Underwater Navigation​

Accurate sensing and mapping are critical for autonomous underwater vehicles operating in obstacle-rich environments. While vision provides high-resolution data, it fails in turbid water, and whereas sonar is robust to turbidity, it suffers from low resolution and elevation ambiguity.

To overcome these limitations, our recent work introduces an opti-acoustic sensor fusion framework that pairs a monocular camera with a stereo sonar to resolve elevation ambiguity and produce fully defined 3D point clouds. These multi-modal points are then fused using a confidence-weighted Gaussian Process Volumetric Mapping framework that prioritizes high-confidence, safety-critical data.

Ultimately, field trials and experimental results validate that this framework successfully captures complex geometries to ensure reliable navigation under degraded sensing conditions.

About the Speaker

Ivana Collado Gonzalez is a Ph.D. candidate at Stevens Institute of Technology, holding an M.S. in Robotics from Stevens and a B.S. in Mechatronics Engineering from Tecnológico de Monterrey, Mexico. Her industry experience includes developing autonomous mobile robots at Xlab Protexa R&D. Ivana’s research focuses on mobile robot exploration, localization, and mapping, specifically advancing marine robotics and perception within complex underwater environments.

Teaching Drones to See What Matters with Reinforcement Learning

Autonomous inspection of industrial environments requires robots to identify and prioritize specific objects of interest, rather than exhaustively exploring their surroundings. This talk presents a deep reinforcement learning framework which enables aerial robots to simultaneously locate, visually inspect semantic targets, and navigate collision-free in unknown environments using only onboard sensors.

The policy generalizes from training on primitive shapes to inspecting complex, real-world structures in real-world settings. The talk also covers a second line of work on active perception, where the the flying agent learns to actively steer its camera sensor during navigation to maximize situational awareness.

Together, these approaches push toward truly autonomous robots that understand not just where to go, but what to look at.

About the Speaker

Grzegorz Malczyk is a PhD candidate at the Autonomous Robots Lab, Norwegian University of Science and Technology (NTNU), where he researches reinforcement learning for autonomous robotic navigation and inspection of industrial environments. He holds an MSc in Robotics, Systems and Control from ETH Zurich and has published across IEEE RA-L, ICRA, and IROS. His work spans semantics-aware path planning, active perception, and sim-to-real transfer for aerial robots.

Gameplay With a Socially Supportive Virtual Robot Enhances Children’s Global Self-Esteem, Peer Relationships, Interest and Engagement

This work investigates whether a socially supportive virtual robot can enhance children's self-esteem and social engagement through game-based interactions. We conducted a month-long study with 23 children in India, where participants played a video game with or without a virtual robot that provided positive reinforcement.
Our results showed that children interacting with the robot demonstrated significant improvements in global self-esteem, friendship quality and quantity, and sustained motivation and enjoyment. These findings highlight the potential of socially supportive virtual robots as tools for promoting children's psychological well-being and social development.

About the Speaker

Devasena Pasupuleti is a researcher in Human-Robot Interaction at the University of Osaka in Japan. Her research focuses on social robotics, conversational AI, and game-based technologies to support and assess children's well-being, learning, and social development. She has authored over 20 publications in leading IEEE and ACM conferences and journals, received multiple Best Presenter and Best Research awards, and is a frequent guest speaker at international research events. Outside her research, she is an author of children's books, reflecting her passion for making robotics accessible and engaging for young audiences.

Safe and Stable Neural Dynamical Systems for Robust Motion Planning

Learning safe and stable robot motions from demonstrations remains a challenge, especially in complex, nonlinear tasks involving dynamic, obstacle-rich environments. In this talk, I will present neural dynamical systems that can help in achieving robust motion plans directly from robot demonstrations.

Moreover, in environments with static algorithms, the framework will yield safe motions with certified confidence bounds.

About the Speaker

Mahathi Anand is a postdoctoral researcher at the Learning Systems and Robotics Lab, Technical University of Germany. She had previously obtained her PhD (Dr. rer. nat.) from LMU Munich, Germany, in 2023. Prior to that, she graduated with a Bachelors in Technology in Electrical and Electronics Engineering from SRM Institute of Science and Technology, India in 2016 and completed her Masters of Technology in Electrical Engineering with specialization in System and Control from Indian Institute of Technology Roorkee, India in 2019.