Pioneering Robotics Research at USC: Showcasing Innovations at ICRA 2025

Researchers at the University of Southern California’s (USC) School of Advanced Computing are leading the way in the field of robotics, presenting groundbreaking findings at the 2025 International Conference on Robotics and Automation (ICRA) in Atlanta, Georgia. This annual event is a platform for the world’s foremost robotics researchers to gather and share their latest innovations and insights, and this year’s conference features a remarkable array of research that pushes the envelope of what is possible in the field.

Cutting-Edge Contributions to Robotics

The USC team’s contributions are varied, demonstrating a broad spectrum of interests and expertise in autonomous systems, human-robot collaboration, biologically-inspired robots, and advanced control mechanisms. Some highlights from their presentations include:

1. DreamDrive: 4D Scene Modeling for Autonomous Vehicles

A significant paper from Jiageng Mao and colleagues introduces DreamDrive, a revolutionary method for generating complex 4D driving scenes derived from real-world data. This system enhances the development of self-driving technology by recreating realistic driving scenarios that can significantly boost the training effectiveness of autonomous vehicle models. The authors leverage generative models to produce dynamic driving videos while ensuring 3D consistency—an achievement that could simplify and enhance the training phase for algorithms used in autonomous navigation.

2. ReMEmbR: Long-Horizon Spatio-Temporal Memory for Robot Navigation

In another compelling presentation, Abrar Anwar and his team unveil ReMEmbR, aimed at enhancing robots’ contextual understanding during navigation. This innovative system allows robots to answer complex queries based on long-term memory, making them more efficient interactors in diverse environments. By employing structured memory approaches, the team demonstrates how ReMEmbR can significantly improve a robot’s ability to understand its surroundings and engage meaningfully with human operators.

3. SAFE-GIL: Safety-Aware Imitation Learning

SAFE-GIL, presented by Yusuf Umut Ciftci and co-authors, focuses on ensuring that robots learn from human behavior safely. This novel approach integrates adversarial disturbances while collecting data, artificially simulating potential errors for more robust training. By emphasizing safety during the learning process, the research aims to mitigate risks and enhance the applicability of behavior-imitating robots in sensitive environments.

The USC School of Advanced Computing

The USC School of Advanced Computing, which encompasses both the Thomas Lord Department of Computer Science and the Ming Hsieh Department of Electrical and Computer Engineering, forms part of the prestigious Viterbi School of Engineering. This multidisciplinary collaboration enriches the research environment, allowing for holistic explorations into advanced computing technologies and robotics.

The school has also been instrumental in organizing key sessions for the conference, showcasing expertise in various cutting-edge topics such as marine robotics, imitation learning, and bimanual manipulation. It’s clear that the impact of USC’s contributions at ICRA reaches far beyond mere presentations; these efforts are helping to shape the future landscape of robotics.

Robotics Innovations on Display

Innovative Research Highlights

Some notable research papers include:

• SMART: Scalable Map Priors for Driving Topology Reasoning by Junjie Ye and his team. This research identifies a critical scalability factor in autonomous driving related to understanding lane connections and traffic elements. Their proposed method leverages standard-definition maps for lane perception, advancing the efficiency of navigation systems.

• A Novel Telelocomotion Framework for Humanoid Robots by An-Chi He et al. This pioneering work integrates a human’s center of mass estimation in the control of humanoid robots, facilitating more effective teleoperation.

• Integrating Field of View in Human-Aware Collaborative Planning by Ya-Chuan Hsu and colleagues. This research addresses the limitations of human perception in collaborative scenarios, creating frameworks that adapt the robots’ actions based on users’ limited visibility—enhancing the efficiency of human-robot cooperation.

Engagement in the Field

The USC researchers are not just presenters but active participants in shaping the discourse around robotics and automation. Their pivotal role in chairing sessions on topics like marine robotics and bimanual manipulation indicates a deep commitment to fostering dialogue and collaboration within the research community.

The breadth and depth of USC’s research at ICRA speaks volumes about the talent and dedication present within the institution. Innovations presented during the conference are not merely academic exercises; they address real-world challenges, equipping robots to operate more safely and effectively alongside humans.

With such prolific contributions to the field, the USC School of Advanced Computing is undoubtedly at the forefront of robotics research, paving the way for future advancements that promise to transform our interactions with machines while enhancing their capabilities in an ever-evolving world.