Franka
PulseAugur coverage of Franka — every cluster mentioning Franka across labs, papers, and developer communities, ranked by signal.
5 day(s) with sentiment data
-
AI Robot Association releases foundation model trained on 16,151 hours of human operation data
The AI Robot Association (AIRoA) has released AIRoA-0, a robot foundation model trained on 16,151 hours of human operation data collected via a handheld gripper called YUBI. This model, built upon Physical Intelligence'…
-
AIRoA-0 system learns complex manipulation from human handheld demos
Researchers have developed AIRoA-0, a bimanual manipulation system trained on over 16,000 hours of human demonstrations recorded using a handheld gripper, without any robot-specific data during training. This system can…
-
RoboColosseum framework reveals weak generalization in robot policies
A new framework called RoboColosseum has been introduced to evaluate the physical intelligence of general-purpose robot policies across diverse embodiments and environments. The framework revealed weak cross-embodiment …
-
逐际动力 FluxVLA Engine V2 streamlines embodied AI development
逐际动力 (LimX Dynamics) has released the second version of its open-source engineering foundation, FluxVLA Engine, designed to streamline the development and deployment of embodied AI. This update enhances capabilities by …
-
New method calibrates AI world model uncertainty for changing policies
Researchers have developed a new method for conformal uncertainty calibration in world models, specifically addressing scenarios where the deployment policy changes while the underlying model remains fixed. This approac…
-
New robot policy recipe ZAP achieves 96.7% success without real-world training
Researchers have developed a new robot control recipe called Zero-real Agent Policy (ZAP) that leverages a pretrained vision-language model. ZAP combines a 27B parameter model for world understanding with supervised fin…
-
New D2R method combats robot policy forgetting using world models
Researchers have developed a new method called Dream to Remember (D2R) to address catastrophic forgetting in generalist robot policies. D2R continually steers a frozen policy towards new tasks while using a pretrained w…
-
New RAFC method improves robot manipulation with reliable future conditioning
Researchers have developed a new method called Reliability-Aware Future Conditioning (RAFC) to improve robot manipulation by making generated future task videos more useful. Traditional methods can be hindered by tempor…
-
Robotic manipulation failure mode diagnosed and solved
Researchers have identified and proposed a solution for a failure mode in long-horizon robotic manipulation tasks, termed Observation-Space Shift (OSS). This issue arises when chaining independently trained skills, caus…
-
LLM-based robots show intent-behavior gap in jailbreaking attempts
Researchers have identified a significant gap between the intended malicious actions and actual harmful behaviors in LLM-based robots, termed the intent-behavior gap. This discrepancy arises because current jailbreak me…
-
RoboCoach uses world models to improve robot skill composition
Researchers have developed RoboCoach, a novel framework that leverages world models to improve robot skill composition. This system uses an RIDI loop (Route-Imagine-Diagnose-Improve) to simulate failures within a world …
-
New framework enhances robotic control by verifying world action model predictions
Researchers have introduced World-Coherent Decoding (WCD), a novel framework designed to enhance the reliability of World Action Models (WAMs) in robotics. WCD operates by treating WAM rollouts as testable hypotheses, s…
-
New SUN Programs framework unifies control and learning for robotic manipulation
Researchers have developed SUN Programs, a novel framework that unifies model-based control and learned policies for complex manipulation tasks. This system, named Kuafu, automatically synthesizes these programs from la…
-
New GCRL method uses dynamic object masks for visual goal representation
Researchers have developed a novel approach for goal-conditioned reinforcement learning (GCRL) that utilizes dynamic object masks as goal representations. This method bypasses the need for privileged state or position i…
-
New MEMENTO framework evolves code-as-policy for embodied AI tasks
Researchers have developed MEMENTO, a novel framework for evolving policies represented as executable code. This memory-guided approach uses a combination of evolutionary methods and local search techniques to improve p…
-
Robots gain 'touch' with 30,000 hours of tactile data from XinZhi Embodied & Fudan
XinZhi Embodied and Fudan University have released a series of technical reports introducing the N0 series models, which aim to enhance embodied intelligence by incorporating tactile data. The project includes a large d…
-
Force Field Intelligence unveils DM0.5 embodied AI model with 150K hours data
Force Field Intelligence has launched its new embodied AI foundation model, DM0.5, which has been trained on 150,000 hours of data. This model aims to address the data bottleneck in embodied AI by integrating real-world…
-
Ant Group releases open-source LingBot-VLA 2.0 for robots · 2 sources tracked
Ant Group's AI division, Ant Group Lingbo, has released LingBot-VLA 2.0, an open-source visual-language-action (VLA) model designed for complex robotic tasks. This new version significantly expands its training data to …
-
Single AI policy controls 20 diverse robot bodies autonomously
A new AI policy, LingBot-VLA 2.0, has demonstrated the ability to control a diverse range of 20 different robot bodies, from single robotic arms to full humanoids, all operating autonomously. This generalist policy was …
-
Uranus: New World Model Acts as Embodied AI Referee and Simulator
Horizon Robotics spin-off Guagua Robot has released Uranus, a world model designed as infrastructure for embodied AI development rather than a direct "brain" for robots. Uranus aims to address two key pain points in the…