MuJoCo
PulseAugur coverage of MuJoCo — every cluster mentioning MuJoCo across labs, papers, and developer communities, ranked by signal.
12 day(s) with sentiment data
-
New framework generates articulated object models from casual videos
Researchers have developed a new framework to reconstruct articulated objects and hand-object interactions from casual monocular RGB videos. This method, called Track, Articulate, Act, does not require depth sensors, mu…
-
Robotic peg insertion enhanced by learned contact dynamics model
Researchers have developed a novel graph neural network model capable of learning contact dynamics for robotic manipulation tasks like peg insertion. This model, trained through self-supervision using only touch and for…
-
AI models compared for wingbeat counting in flapping-wing vehicles
Researchers have evaluated three types of temporal models—convolutional, spiking, and attention-based—for counting wingbeats in flapping-wing vehicles using optical flow data. The study, conducted in the MuJoCo simulati…
-
New AI framework uses human expertise for precise robotic construction assembly
Researchers have developed a new framework for interactive reinforcement learning that leverages human expertise to improve the precision of robotic assembly in industrialized construction. This system aims to convert t…
-
New benchmark tests LLM physical reasoning via interactive physics simulation
Researchers have introduced PhysMent, a new benchmark designed to evaluate the physical reasoning capabilities of large language models (LLMs) through interactive experimentation. Unlike static benchmarks, PhysMent requ…
-
New Zonal RL-RRT algorithm boosts path-planning efficiency
Researchers have developed a new path-planning algorithm called Zonal RL-RRT, which significantly improves efficiency and success rates in complex environments. This algorithm partitions maps into zones using a k-d tree…
-
Guide to Imitation Learning for Robotics Released
This article provides a guide to implementing imitation learning (IL) for robotics, focusing on vision-based policies trained from scratch. It contrasts IL with classic explicit policies and reinforcement learning, high…
-
New Ostrich simulator enables faster, more memory-efficient differentiable dynamics
Researchers have developed Ostrich, a novel GPU-accelerated rigid-body simulator designed for differentiable dynamics. Ostrich addresses limitations in existing simulators by enabling gradient-based optimization through…
-
Qunhe Tech, Nvidia, Intel, Zhejiang U. tackle physical AI infrastructure
Qunhe Technology, in collaboration with industry giants like Nvidia and Intel, and academic institutions such as Zhejiang University, has published three papers addressing key infrastructure challenges for physical AI. …
-
New visuotactile world model boosts robotic lifting success rates
Researchers have developed a compact visuotactile world model that improves robotic manipulation by integrating visual and tactile data for more accurate predictions. This model enhances force-constrained control, signi…
-
Axis Robotics launches browser-based data engine for robot manipulation research
Axis Robotics has introduced AXIS, a novel browser-based data engine designed to accelerate robot manipulation research. This system allows for continuous data collection through a web interface, with backend GPUs handl…
-
Sparse world models improve AI object manipulation accuracy
Researchers have developed a new approach to world models in AI, focusing on modeling changes rather than predicting entire states. This sparse, residual world model demonstrated significantly higher accuracy in predict…
-
New SEBA framework enables sample-efficient black-box attacks on visual RL
Researchers have developed SEBA, a new framework designed to efficiently launch black-box adversarial attacks against visual reinforcement learning agents. This method integrates a shadow Q model to estimate rewards und…
-
New method extracts biomechanical pose from 3D body models
Researchers have developed a new method to extract biomechanically accurate joint angles from single RGB images, addressing a limitation in current 3D body recovery techniques. This approach extends the SAM 3D Body foun…
-
MirroS unveils Code-as-World for video-to-physics simulation
MirroS has developed Code-as-World, a system that transforms real-world videos into executable MuJoCo physics simulations. This system employs an agentic loop to create and verify these simulated environments, which are…
-
Open-source robot duck trained with AI available for pre-order
Pollen Robotics and Hugging Face have collaborated to create an open-source bipedal robot duck, named MicroDuck. This 25cm robot features 15 motors, a camera, LiDAR, and a grasping beak. It is trained using reinforcemen…
-
New research explores robot morphology and control learning optimization
Two new research papers explore advancements in robot morphology and control learning. The first paper introduces AdaControl, a method that optimizes robot morphology and control learning simultaneously, reducing comput…
-
Chinese team DeepSoma aims for detailed brain simulation and real-world integration
Chinese startup Zhiyue Kongjian Intelligence has launched DeepSoma, a platform aiming to simulate detailed biological neuron models and integrate them with real-world environments and diverse physical bodies. Unlike Eon…
-
FEAGINE unveils flexible robot bodies and cross-embodiment AI model
FEAGINE, a robotics company founded by former DJI engineer Peng Rui, is developing a new approach to embodied AI that moves beyond the humanoid form. Instead of replicating human bodies, FEAGINE is designing flexible, b…
-
New DClamp-PPO algorithm enhances reinforcement learning by penalizing 'wrong' direction updates
Researchers have introduced Directional-Clamp PPO (DClamp-PPO), a novel algorithm designed to enhance the performance of Proximal Policy Optimization (PPO) in deep reinforcement learning. DClamp-PPO addresses a key limi…