LSY @ ICRA 2026

Why It Matters

These advances address core challenges in reliability, safety, and scalability of autonomous robots, directly influencing industrial automation, logistics, and aerial services. By uniting AI‑driven perception with control theory, the research accelerates deployment of trustworthy, cooperative robots in complex real‑world environments.

Key Takeaways

• Robust nonprehensile transport improves manipulation reliability
• Diffusion-based crowd navigation reduces collision risk
• Safety filters enhance diffusion policy trustworthiness
• Human-prediction MPC boosts collaborative robot performance
• SwarmGPT merges LLM reasoning with multi-agent planning

Pulse Analysis

ICRA remains the premier venue for unveiling cutting‑edge robotics research, and this year’s lineup reflects a decisive shift toward AI‑augmented control. Papers on robust nonprehensile object transportation and sensor scheduling demonstrate how probabilistic reasoning can compensate for hardware limitations, delivering higher success rates in unstructured settings. Meanwhile, diffusion‑based crowd navigation and safety‑filtered diffusion policies illustrate the growing confidence in generative models to predict safe trajectories, a trend that could replace handcrafted planners in dense urban environments.

The lab’s human‑prediction MPC and range‑aided cooperative MAV navigation address the perennial challenge of integrating intent and perception into real‑time decision making. By forecasting human motion, the MPC framework enables smoother human‑robot collaboration, essential for manufacturing and service robots. The MAV study leverages inter‑vehicle ranging to maintain formation despite noisy sensors, showcasing scalable solutions for aerial swarms used in inspection, delivery, and disaster response. Together, these works underscore a holistic approach where estimation, prediction, and control co‑evolve.

From a market perspective, these contributions lower barriers to commercial adoption of autonomous systems. Safety filters and LLM‑driven swarm planning provide the transparency and adaptability demanded by regulators and end‑users alike. As industries seek to automate logistics, warehouse operations, and aerial logistics, the presented technologies promise reduced downtime, enhanced safety, and new business models centered on collaborative robot fleets. The convergence of diffusion AI, model‑predictive control, and large language models signals a new era of intelligent, cooperative robotics poised for rapid scaling.

Original Description

Our lab is excited to present 7 papers at ICRA 2026:

- Robust Nonprehensile Object Transportation: https://arxiv.org/abs/2411.07079

- SICNav-Diffusion Crowd Navigation: https://arxiv.org/abs/2503.08858

- Safety Filtering for Diffusion Policies: https://arxiv.org/abs/2511.06385

- SM²ITH Human-Prediction MPC: https://arxiv.org/abs/2511.17798

- SwarmGPT (LLMs + Swarm Planning): https://arxiv.org/abs/2412.08428

- Sensor Scheduling & Noise for Estimation: https://doi.org/10.1109/LRA.2025.3570948

- Range-Aided Cooperative MAV Nav: https://ieeexplore.ieee.org/document/11235959