A video model that respects physics: Chinese team open-sources a foundation for training robots
A Chinese team has open-sourced a video model that does not violate the laws of physics. The problem: AI videos often show the impossible, such as water behaving like jelly or objects opening without any contact. For human viewers it is a visual glitch, but for robots that learn from these videos it is a serious issue: they memorize incorrect physics and make mistakes when performing real actions. The new model corrects this and serves as a base for properly training robot control systems.
AI-processed from Jiqizhixin (机器之心); edited by Hamidun News
On July 9, 2026, a Chinese team open-sourced a video model optimized specifically for teaching robots real movements and actions. The work addresses a critical problem: today's video generators often produce physically impossible scenes that mislead robot training systems.
The problem: physics violations in AI video
Watch a video made by a neural network and you will inevitably spot errors that break physical laws. Water pours out of a cup but hangs in the air like transparent jelly instead of falling under gravity. A hand reaches toward a drawer without even touching it, yet the drawer starts sliding open, as if some miraculous force were acting at a distance. For an ordinary person watching for entertainment, this is just a film goof, a visual artifact, a minor inaccuracy that is easy to overlook.
- Typical physical errors: violations of gravity, incorrect fluid physics, impossible rigid-body mechanics
- For a human viewer: just a visual defect, not critical to understanding
- For a robot or an AI training system: incorrect, misleading information
Why this matters for robotics
The problem becomes serious when these videos are used to train large models that then control robots or help them learn correct actions. The model learns not only from correct movements but also from every physics violation in the source data.
Imagine a robot watching a video in which a hand opens a drawer from a distance without touching it. The system memorizes this pattern as a normal action. Later, when the robot is asked to open a real drawer, it copies the pattern: it reaches out close to the drawer but does not touch it. The result: the drawer will not open, and the robot makes an error. The system cannot function in the real physical world, because its model of interactions is built on physically impossible examples.
The solution: a video model from a Chinese team
The open-source video model was developed with physical and geometric correctness requirements in mind. It generates videos that obey real physical laws: objects fall correctly, liquids flow naturally, and contacts between objects are real.
The model serves as a base for training systems that control robots or help them learn correct movements. Systems that use it receive accurate information about how the physical world actually works.
What this means
The release shows that practical use of AI video in robotics requires specialized, task-tuned models. General-purpose video generators cannot simply be used as they are; versions that respect physics and real-world laws are needed. This is a critical step toward smarter, more reliable and safer robots that can operate in the real world without errors.
- Nov 1, 2026Pika Labs releases open video model H3 Flex: 6 steps instead of 50
- Nov 1, 2026Google's StreamVibe video agent beats other models by 10.7% on response timing and quality
- Oct 31, 2026AI cartoon series: a 24-second episode built ten ways, the best cost 162 credits
- Oct 31, 2026Anthropic: Fold Subagents Back Into a Single Loop — One Customer's Tasks Ran Twice as Fast
Key Questions Answered
What are the key details on The problem: physics violations in AI video?
Watch a video made by a neural network and you will inevitably spot errors that break physical laws. Water pours out of a cup but hangs in the air like transparent jelly instead of falling under gravity. A hand reaches toward a drawer without even touching it, yet the drawer starts sliding open, as if some miraculous force were acting at a distance. For an ordinary person watching for entertainment, this is just a film goof, a visual artifact, a minor inaccuracy that is easy to overlook. Typical physical errors: violations of gravity, incorrect fluid physics, impossible rig
Why this matters for robotics?
The problem becomes serious when these videos are used to train large models that then control robots or help them learn correct actions. The model learns not only from correct movements but also from every physics violation in the source data. Imagine a robot watching a video in which a hand opens a drawer from a distance without touching it. The system memorizes this pattern as a normal action. Later, when the robot is asked to open a real drawer, it copies the pattern: it reaches out close to the drawer but does not touch it. The result: the drawer will not open, and the
What are the key details on The solution: a video model from a Chinese team?
The open-source video model was developed with physical and geometric correctness requirements in mind. It generates videos that obey real physical laws: objects fall correctly, liquids flow naturally, and contacts between objects are real. The model serves as a base for training systems that control robots or help them learn correct movements. Systems that use it receive accurate information about how the physical world actually works.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.