Agibot has officially launched its AGILE 2.0 integrated vision-and-control model, marking a significant advance in enabling robots to "see, think, and move" simultaneously, achieving a level of natural and fluid motion comparable to human beings.

In a departure from conventional technical demonstrations, the company showcased this breakthrough through a creative short film titled "Acrobat BOT," which vividly demonstrates how robots can genuinely use their vision to execute complex, dynamic tasks in real-time.

Visual Feats in the Robot Circus

The video opens with a high-wire act: the Lingxi X2 robot precisely locates the center of a flaming hoop, leaps effortlessly, and clears it with ease. It then proceeds through a series of physically demanding routines, including diabolo spinning, three-person jump rope, leapfrog, paired high-platform climbs, three-ball juggling, and collaborative box carrying.

These tasks all tackle the industry's central challenge: enabling robots to first observe with their "eyes," then plan and act. The entire film uses multiple vision-driven, high-difficulty collaborative tasks to rigorously demonstrate the breakthrough capabilities of AGILE 2.0's end-to-end vision-based control architecture.

From Seeing to Acting: Solving the Latency Challenge

For robots, seeing is not the hard part; reacting to what they see in real-time is. In humans, eyes and body are naturally synchronized; the eyes capture changes, the brain predicts delays, and limbs adjust smoothly. Robots, however, operate as separate subsystems. There's an inherent time lag between when a camera captures an image and when the AI calculates a target position. Since a robot's motors react in milliseconds, using delayed vision data to guide movement often results in jitter, shaking, and instability.

With AGILE 1.0, Agibot established a preliminary link between perception and action, but mobile manipulation was limited to parallel processing, lacking whole-body coordination and real-time coupling. Most industry approaches remain a simple layering of movement and manipulation, without a unified dynamic framework, leading to instability and failure in tasks involving precise balance, rapid interaction, or fine manipulation.

Core Innovations: Unified Vision-to-Action Architecture

AGILE 2.0 builds upon its predecessor by achieving deep integration across the entire pipeline. This includes environmental observation, terrain mapping, dynamic traversability analysis, full-body motion control, contact state switching, and precise end-effector manipulation. The model operates as a true end-to-end system, with no intermediate modules or human-devised step divisions.

The most compelling demonstration is the Lingxi X2 balancing on an exercise ball while rolling it forward. This feat, resembling a skilled circus performer, highlights extraordinary dynamic balance and motor control. This marks the first time a bipedal humanoid robot has autonomously accomplished ball balancing, underscoring the industry-defining progress of AGILE 2.0.

In another demonstration, two robots swing a long rope rhythmically while a third uses vision to time its entry and exit from the jumping sequence. In a "box moving" scenario, one robot picks up a box, and another observes, approaches, knocks it down—as if to say "let me show you how it's done"—then lifts it and carefully stacks it onto another. These examples showcase whole-body loco-manipulation, enabling simultaneous precise handling during dynamic movement while maintaining overall stability.

Real-World Impact: Scaling Deployment in Complex Environments

The core technical achievement of AGILE 2.0 is its closed-loop, vision-based control system. In scenarios with rapid environmental changes and shifting obstacles, it enables real-time visual reasoning and dynamic constraint optimization. For tasks involving two or more robots, it establishes cross-robot state awareness and synchronized behavioral coordination.

Furthermore, the model incorporates safety measures for human-robot coexistence: if a person unexpectedly enters the workspace, the system implements protective responses and autonomous recovery. By breaking through the limitations of structured environments, AGILE 2.0 is designed to operate in dynamic, unpredictable settings, reducing the need for heavy customization. This advancement paves the way for embodied intelligence to reach scalable, practical, and commercially viable production at scale.