$π\mathbf{R}^2$: Reactive Real-time Flow Policies
Researchers introduced Pi R2, a method that allows generalist robot manipulation policies to react to sensory input in real time, overcoming previous latency issues.
- Pi R2 addresses the latency bottleneck in perception-to-action pipelines for robotics.
- It enables closed-loop control for action-chunking flow policies by allowing frequent replanning.
- The method allows robots to react to sensory input during execution, improving performance in dynamic environments.
Generalist manipulation policies often rely on action-chunking, where a sequence of actions is predicted at once. This approach runs open-loop, meaning the robot cannot adjust its course based on new sensory information that arrives while the action sequence is being executed.
The primary barrier to fixing this is latency. The perception-to-action pipeline, which involves large pretrained backbones and multiple denoising steps, is computationally heavy. This makes frequent replanning impossible, leaving the robot unable to react to dynamic changes in its environment.
The paper presents Pi R2, a technique designed to bridge this gap. It optimizes the pipeline to allow for frequent replanning, effectively restoring reactivity to these policies. This enables robots to perform closed-loop control in real time, making them suitable for dynamic and unpredictable tasks.
Provides a new method to implement responsive real-time control in robotic systems.
Enables more capable robots for complex, dynamic real-world applications.
Signals progress in overcoming technical hurdles for generalist robotics commercialization.
- Action-chunking
- Predicting a sequence of future actions at once rather than one step at a time.
- Open-loop
- A control system that executes actions without using feedback from sensors to adjust during execution.
Pictura: Perspective-View Self-Play at Scale for Driving
RoboticsNVIDIA Cosmos-H-Dreams: Bringing Real-Time Generative Simulation to Surgical Robotics
RoboticsThis is the world's most advanced robotic servicing satellite—that we know about
RoboticsRobot snakes searched for Venezuela earthquake survivors in collapsed buildings
RoboticsHyundai claims humanoid robot plan is not part of talks with striking workers
AI ResearchAnthropic Mythos Preview Raises the Stakes for AI-Assisted Cryptography Research
Anthropic has introduced the Claude Mythos Preview and Project Glasswing, significant developments in AI-assisted cryptography research.
New research framework aims to assess and track clinical AI models - Healthcare IT News
Researchers introduce a new framework to assess and track clinical AI models, aiming to improve healthcare outcomes.
AI ResearchOpenAI’s GPT-5 Science Report Puts Human Stewardship at the Center of AI Research
OpenAI has published a report on GPT-5, highlighting the importance of human stewardship in AI research.
Pusan National University Study Rethinks How Artificial Intelligence Supports Investment Decisions - PR Newswire
A study by Pusan National University explores how artificial intelligence can support investment decisions. The research aims to improve the accuracy of investment predictions using AI.
AI ToolsOpenWorker: Andrew Ng's Local-First AI Coworker, Explained for Developers
OpenWorker, a local AI coworker developed by Andrew Ng, has been shipped in late July 2026. It is an MIT-licensed tool that runs on users' own machines.
Why AI-driven enterprises are the future of entrepreneurship - MIT Sloan
MIT Sloan discusses the role of AI in shaping the future of entrepreneurship, highlighting its potential to drive innovation. AI-driven enterprises are expected to revolutionize the industry.