Autonomous Grand Challenge 2026

High-dynamic ball sports such as table tennis pose extreme challenges for humanoid robots, requiring tight integration of high-speed perception, real-time decision-making, and agile whole-body control. This challenge uses table tennis as a representative task to systematically evaluate whether humanoid robots can achieve human-like performance in fast-paced, adversarial sports scenarios. The competition is designed in strict accordance with official table tennis rules, balancing fairness, realism, and technical difficulty. Participants are expected to develop complete perception-planning-control systems capable of handling rapid ball motion, precise contact timing, and dynamic opponent behaviors.

Deformable Object Manipulation, particularly the handling of textiles such as clothing, remains one of the most fundamental and unsolved challenges in robotics. Unlike rigid objects, deformable materials exhibit effectively infinite degrees of freedom and highly non-linear dynamics, making perception, modeling, and control intrinsically difficult. Despite recent progress, many existing approaches rely heavily on simplified simulation environments or reduce manipulation to isolated pick-and-place primitives, which fall far short of capturing the complex, continuous, and history-dependent dynamics of real-world. The challenge is designed as a scientific instrument to assess a system's full-stack capability. Participants are tasked with developing end-to-end solutions for flat folding garments. This task requires the coordinated integration of computer vision, control theory, and learning-based methods such as deep reinforcement learning, under realistic physical constraints.

Dexterous manipulation in unstructured environments demands both high-dimensional control and rich sensory feedback. This challenge advances research on High-DoF anthropomorphic hands with multimodal sensing - proprioception, tactile, and binocular vision, to enable robust, contact-rich manipulation. Participants will be evaluated on a suite of standardized manipulation tasks that test precision, robustness, and generalization.

Contact-rich manipulation, such as peg-in-hole, tool use, and mechanical assembly, involves fine-grained physical interactions where vision alone is insufficient to capture critical local contact states, real-time forces, and subtle deformations. The challenge aims to systematically advance visuo-tactile learning methods for real-world high-precision manipulation and to clarify the role of touch in practical assembly and tool-based operations. Submitted approaches will be evaluated on their performance across multiple manipulation tasks, with a particular emphasis on effectiveness, efficiency, and generalization. This challenge aims to systematically assess the role of tactile perception in robotic manipulation and to advance the development of robust, generalizable visuo-tactile learning methods for real-world physical interaction.

This challenge targets vcontact-rich household and tool-use behaviors that require precise coordination between two hands, continuous control, and robust perception. Participants develop policies that map multi-view visual observations and natural-language instructions to joint-space control, with evaluation designed to reflect both task completion and the stability of physical interaction.