Franka Robotics

Proud to see our robot arms featured in a video by Jason Liu (Developer Experience Engineer, Codex team OpenAI) about OpenAI's mission to build AI that benefits all of humanity. We're pleased to have such an outstanding customer. 🤝

👋🏻 Meet us at IJCAI-ECAI 2026! From 15–21 August, we’ll join the AI and robotics community in Bremen - a gathering of researchers exploring the next generation of intelligent systems. From manipulation and imitation learning to Physical AI, the conference brings together experts shaping how robots can learn, reason, and operate in complex real-world environments. That’s why we’re bringing our FR3 Duo - the reference platform for Physical AI that unifies teleoperation, high-quality data collection, and policy execution in one integrated setup. It enables rich real-world interaction data collection, including tactile information, and the direct evaluation and deployment of learned policies on the robot. FR3 Duo connects AI development with physical interaction, enabling to bring learned skills from controlled environments into real-world scenarios. We look forward to exchanging ideas and connecting with the community in Bremen! 📩 Book a meeting with our on-site team: research@franka.de #IJCAI2026 #Bremen #AI #Robotics #FR3Duo #PhysicalAI #FrankaRobotics

HABIT: a dataset for human-robot collaboration 🤝 When a robot works next to a person, most of what matters happens in the small stuff: stepping aside a bit early, noticing a raised hand before it becomes an instruction. HABIT, a new dataset from the Config team, was built to capture exactly that - across three interaction modes:   - Coordinated collaboration on shared tasks - Safe yielding in shared spaces - Recognizing human gestures and intent   The dataset was collected using two Franka Research 3 robots and includes over 10,000 episodes and more than 160 hours of interaction data. It was evaluated using several Vision-Language-Action models.   The results suggest that progress in robot learning depends not only on more data, but on data deliberately collected to reflect real human-robot interaction.   Congratulations to the HABIT team at Config and 한국과학기술원(KAIST) 부설 한국과학영재학교!   📄 Paper: arxiv.org/abs/2606.31682   Building datasets like HABIT requires robust data collection workflows. If you're curious about that process, our open-source data collection suite - LABS offers a practical starting point → www.franka.de/labs   #Robotics #RobotLearning #PhysicalAI #HumanRobotInteraction

Physical AI is moving fast - but how do you turn the hype into real-world deployment? 📢 Hear from industry leaders from robotics, AI, and manufacturing to explore the different approaches, what’s working today, and what it takes to make Physical AI deliver real value. Join Robotiq CEO Samuel Bouchard and industry experts from Franka Robotics, Teradyne, NVIDIA Robotics, and Robotiq for a candid discussion on: 🔹 What Physical AI can actually do today 🔹 Where the technology still falls short 🔹 What it takes to move from impressive demos to real manufacturing value 🔹 What manufacturing teams should consider before investing in Physical AI 📅 Thursday, August 6 · 11 AM ET 🎥 Can't attend live? Register anyway and Robotiq team send you the recording. 👉 Register: https://lnkd.in/gf9TZqCJ #PhysicalAI #Robotics #Manufacturing #AI

A major step forward for embodied AI and intelligent robotics. 🦾 Google DeepMind introduced Gemini Robotics ER 2, showcasing how advanced reasoning models enable robots to understand complex tasks, collaborate seamlessly, and operate more effectively in the physical world. We're proud to see Franka FR3 Duo featured in the multi-robot collaboration demonstration, highlighting how heterogeneous robots can work together to accomplish tasks that would be difficult for a single robot alone. This milestone also reflects a broader journey. As an Agile Robots SE company, we're proud that Google DeepMind is our strategic partner, working together to bring Gemini Robotics foundation models into real-world robotic systems and accelerate the development of intelligent, adaptable automation. At Franka Robotics, we believe the future belongs to robots that don't just execute commands - they understand context, collaborate intelligently, and continuously learn from the physical world. 🎥 Watch the video to see embodied AI in action and how Franka FR3 Duo contributes to the next generation of intelligent robotics. #FrankaRobotics #AgileRobots #GoogleDeepMind #GeminiRobotics #EmbodiedAI #PhysicalAI 

A smarter way to slice 🍣 Manipulating deformable objects remains one of the biggest challenges in robotics. Unlike rigid objects, materials such as food or biological tissues vary in shape, are difficult to grasp, and behave unpredictably during interaction. The Sashimi-Bot project tackles this challenge with three Franka Research 3 robots, combining visual and tactile sensing, deep reinforcement learning, and multi-robot coordination to autonomously prepare sashimi. This demonstrates how robots can adapt to complex, real-world manipulation tasks and opens new possibilities for applications involving delicate and deformable objects. The Franka Research 3, with its high dexterity, precision, and compliance, enables this type of advanced contact-rich research. Featured in Nature Research Highlights of the Nature Portfolio journal, the work is a great example of international collaboration between SINTEF Ocean, Norwegian University of Science and Technology (NTNU), NMBU - Norwegian University of Life Sciences, Massachusetts Institute of Technology, Inria, and QUT (Queensland University of Technology). Congratulations to Ekrem Misimi and the entire Sashimi-Bot team on this achievement! We are proud to see Franka Research 3 robots supporting research that continues to expand the capabilities of robotic manipulation for dynamic real-world environments. 📄Learn more about the project: https://lnkd.in/gEwiVK8p Find out more about the reference platform for Robotics & AI: https://lnkd.in/g_ck-3rg #Robotics #PhysicalIntelligence #RobotLearning #RoboticManipulation #FrankaResearch3

💡 What you need to scale robots for the real world?👇🏼 The next challenge for Physical AI is not only teaching robots new skills - it is enabling them to learn from failure and keep improving once they interact with the real world. A new research from ETH Zürich, Technical University of Munich, Microsoft, and Munich Center for Machine Learning explores how robots can use knowledge from human videos to better understand their own actions, recover from failures, and continuously improve their manipulation skills. Across seven real-world manipulation tasks and multiple robot embodiments, the approach increased success rates from 40% to 81% - showing the potential of combining human knowledge with robot experience. With its precision, force sensitivity, and open research capabilities, Franka robots provide the foundation needed to develop Physical AI. Researchers can demonstrate skills, generate datasets, including tactile data, and execute learned policies through low-level 1 kHz control via Franka Control Interface (FCI) - supporting the full development cycle from learning to deployment. 👏🏼Congratulations to the teams behind this research! Find out more about this project and full paper from the link in the Comments! 👇 Discover how we unlock Physical AI at scale: https://lnkd.in/d9a8UHuG #PhysicalAI #RobotLearning #RobotManipulation #EmbodiedAI #RoboticsResearch

Teaching robots to feel, before they ever touch the real world. 🤖 Humans naturally adapt their grip when objects move, slip, or require delicate handling. For robots, mastering these contact-rich interactions remains a major challenge. HydroShear, developed by Amazon Science and the University of Michigan , tackles this challenge with a physics-based tactile simulator that captures how forces evolve during contact — allowing robots to learn complex manipulation skills in simulation. The result? Skills learned entirely in simulation transferred directly to a real Franka Research 3 robot equipped with GelSight Mini sensors, without additional training or fine-tuning, achieving a 93% average success rate across four manipulation tasks. With its open research interface, real-time control capabilities, software ecosystem, and flexible sensor integration, Franka Research 3 enables researchers to develop and transfer new manipulation approaches from simulation into the real world. A great example of how simulation, tactile sensing, and our FR3 robot are accelerating progress in physical AI. 👏🏼 Congratulations to the teams at Amazon Science and the University of Michigan on this impressive work. Explore, build, and push the boundaries of robot learning with Franka Research 3 and LABS: https://franka.de/labs #PhysicalAI #EmbodiedAI #Robotics #TactileSensing #Sim2Real #FrankaResearch3

Teaching robots to think ahead. 🤖 To bring AI into the physical world, robots need more than just vision. They need to predict, reason, and act. Robbyant’s LingBot-VA 2.0 explores a new approach: an embodied-native world-action model pretrained from scratch for physical interaction. Using our Franka Research 3 (FR3) robots, the team demonstrates how this approach enables robots to tackle complex manipulation tasks, including high-precision chip grasping, few-shot conveyor belt assembly, and desktop cleaning with real-time foresight reasoning. By predicting future world states while acting, LingBot-VA 2.0 helps move robots toward more adaptive behavior, faster learning, and more complex task execution. With its precision, torque sensing at every joint, and safe interaction capabilities, FR3 provides an ideal platform for advancing this kind of Physical AI research. 🎉Impressive work by the Robbyant team on pushing the boundaries of embodied AI. We’re excited to see where this research leads next. 👉🏼Learn more about LingBot-VA 2.0: https://lnkd.in/eEs5y8V4 #PhysicalAI #EmbodiedAI #RobotLearning #Robotics

"Intelligence must exist where the work happens" At Toshiba Europe Ltd's Cambridge Research Laboratory, researchers are exploring this frontier by building the foundations of Physical AI — robotic systems that can perceive, reason, and act in the real world. We’re proud to see the Franka Research 3 robot supporting this research journey, enabling the team to explore how robots can move beyond predefined motions and adapt to the complexity of real-world environments. With an open research ecosystem and low-level control access, FR3 supports the full robot learning workflow: → Demonstrating manipulation skills → Generating high-quality datasets, including tactile data → Deploying and benchmarking learned policies → Exploring precise control through 1 kHz access via Franka Control Interface (FCI) This helps researchers bridge the gap between AI models and real-world robotic action. 👏🏼 Great work by the Toshiba Europe Ltd Cambridge Research Laboratory team, leveraging the Franka Research 3 robot to push the boundaries of Physical AI research. 🔎Discover more examples of how researchers are using Franka robots to advance Physical AI: https://lnkd.in/d9a8UHuG #PhysicalAI #EmbodiedAI #RobotLearning #FrankaRobotics #RoboticsResearch #AI