“`html
A team of researchers from Carnegie Mellon University and the Bosch Center for AI have developed a new AI system called HTD (Humanoid Transformer with Touch Dreaming).
- Their research focuses on improving how humanoid robots can perform tasks that require dexterous whole-body manipulation in real-world scenarios.
- In five specific tasks—such as inserting objects, organizing books, folding towels, scooping cat litter, and serving tea—HTD demonstrated a 90.9% higher success rate compared to previous methods.
- Interestingly, the effectiveness of HTD was attributed not just to adding touch signals but also in predicting tactile signals in latent space rather than directly from raw data, which provided an additional 30% improvement in performance.
“`
### Takeaways
– HTD significantly enhances humanoid robots’ ability to perform complex tasks involving dexterous manipulation.
– The key innovation lies in the use of latent space predictions for tactile signals, leading to substantial improvements over raw input methods.
– This advancement could lead to more efficient and reliable robotic assistance in various real-world applications.
Source Read original →
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
