3D CT reconstruction, CAD generation, and dexterous robotic hands push boundaries
ScienceToStartup Editorial
AXON unveiled a notable advance in medical imaging by reconstructing 3D CT volumes directly from 2D X-rays, improving diagnostic accessibility. Meanwhile, CADSmith introduced a multi-agent pipeline for generating CAD models with programmatic geometric validation, achieving a 100% execution rate. Ruka-v2, a tendon-driven robotic hand, enhances dexterity with added wrist mobility, showcasing significant improvements in teleoperated tasks.
🩻 Medical Imaging
The Rundown
AXON just launched a multi-stage diffusion-based framework that reconstructs high-fidelity 3D CT volumes from 2D X-rays. This innovative approach addresses the limitations of traditional CT scans, such as high radiation exposure and cost. AXON employs a coarse-to-fine strategy, achieving an 11.9% improvement in PSNR and an 11.0% increase in SSIM over current best methods. By supporting bi-planar X-ray input, AXON effectively mitigates depth ambiguities, enhancing diagnostic capabilities across various clinical settings.
The details
Why it matters
AXON's approach significantly enhances diagnostic accessibility, enabling healthcare providers to leverage existing 2D X-ray data for more comprehensive analysis. This could lead to earlier interventions and improved patient outcomes, particularly in resource-limited settings.
📐 Text-to-CAD tools
The Rundown
CADSmith has introduced a multi-agent pipeline that generates CAD models from natural language prompts, enhancing the reliability of CAD generation. The system employs two nested correction loops, achieving a 100% execution rate and improving the median F1 score from 0.9707 to 0.9846. By integrating programmatic geometric validation, CADSmith ensures precision in dimensions and solid validity, making it a robust tool for designers and engineers alike.
The details
Why it matters
CADSmith's advancements in reliable CAD generation empower designers to create complex models with confidence. As industries increasingly adopt digital design tools, this system could streamline workflows and reduce errors, ultimately enhancing productivity.
🤖 Robotics
The Rundown
Ruka-v2 has been unveiled as an open-source, tendon-driven humanoid hand, improving upon its predecessor with added wrist mobility and finger abduction capabilities. This new design allows for more natural manipulation in confined spaces, achieving a 51.3% reduction in task completion time and a 21.2% increase in success rates during user studies. Ruka-v2's versatility makes it suitable for a wide range of robotic applications, from teleoperation to autonomous learning.
The details
Why it matters
Ruka-v2's enhancements in dexterity and adaptability position it as a key player in the development of advanced robotic systems. Its open-source nature encourages collaboration, potentially accelerating innovation in robotic applications across industries.
Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.
Community Insight in 👥
“'I'm a robotics enthusiast and recently built Ruka-v2. The dexterity is impressive! I used it for a project involving precise manipulation of small objects, and it significantly improved my workflow. The community support around the design files is fantastic, making it easy to troubleshoot and innovate.'”
A library for NLP, vision, and multimodal tasks with pre-trained models.
Built to make you extraordinarily productive, Cursor is the best way to code with AI.
A platform for tracking experiments, datasets, and model performance.
A flexible framework for building and training ML models.
An open platform for managing the full ML lifecycle.
A framework for building applications powered by LLMs.
AXON improves diagnostic capabilities using existing 2D X-rays.
CADSmith enhances CAD generation accuracy through iterative refinement.
Ruka-v2's open-source model promotes collaborative robotics development.
AXON's performance validated across public and external datasets.
CADSmith maintains a current database for evolving CAD libraries.
Apr 3
OmniMem's multimodal memory, ORCA's LLM calibration, and VRUD's urban traffic dataset
Apr 1
Adaptive quantization, agentic image generation, and long video frameworks redefine capabilities
Mar 31
Adaptive data types, agentic image generation, and on-device models reshape AI landscape