
Purdue's McVAMP Plans Humanoid Robot Motion 1000x Faster on a CPU
A new SIMD-accelerated planner from Purdue's CoMMA Lab drops humanoid constrained motion planning from tens of seconds to milliseconds, enabling reactive whole-body control.
20 stories

A new SIMD-accelerated planner from Purdue's CoMMA Lab drops humanoid constrained motion planning from tens of seconds to milliseconds, enabling reactive whole-body control.

A new sampling strategy uses Pontryagin's Maximum Principle to concentrate training data near safety-critical states, cutting failures and hitting 3ms inference for real-time safety filtering.

A second, delayed loudspeaker adds frequency-selective energy to the reverberant field, correcting both timbre and spatial perception without listeners noticing it.

A two-stage training recipe teaches deepfake detectors to model natural prosody first, cutting error rates by more than half on emotional spoofing attacks.

PLanet is a domain-specific language that treats experimental designs as matrix constraints, catching hidden assumptions before you run a single trial.

A visualization technique from Zhejiang University builds nested entity graphs sentence by sentence, cutting question response time and doubling answer quality in user studies.

Carnegie Mellon researchers turn the eye's cornea into a passive world-facing camera, using tiny retroreflective tags to identify objects and estimate their 3D pose from glint patterns alone.

A qualitative CSCW study reframes older Korean immigrants' digital avoidance as strategic refusal, not skill deficits, with implications for inclusive design.

A multi-agent LLM system that gently steers travelers toward greener choices using counterfactual explanations, without overriding user intent.

Meta and UCLA researchers show that 19 popular re-rankers stubbornly prefer old, detailed documents over newer ones, and offer a fix without retraining.

A Caltech astronomer built a browser and VS Code extension that turns rough placeholder cite keys into real BibTeX entries, without leaving Overleaf.

A new benchmark tests 14 OCR models across 158 Unicode scripts — and even the best frontier model fails on 92% of low-resource scripts

Researchers reverse-engineered Claude Code's 512K-line TypeScript codebase and found that only 1.6% is AI logic — the rest is a fortress of deterministic infrastructure built to manage the model
A Tsinghua paper dissects why on-policy distillation fails, revealing two hard rules and two practical fixes that every LLM post-training pipeline should know

Microsoft's screen parser has crossed 25k stars and just added a YOLOv9-E detector, sharpening vision-only GUI agents built on top of any LLM.

fli gives Claude and any MCP-compatible AI direct access to Google Flights data — no scraping, no API key, fully open source

Claude Code Viewer is an open-source web client that turns your local Claude Code sessions into a browsable, mobile-friendly workspace with Git tooling and remote access.

A community GGUF quant of Google's Gemma 4 26B MoE strips safety guardrails and hits 89 tokens per second on Apple Silicon.

Every-Embodied is a free, open-source course that takes you from basic Python to training and deploying state-of-the-art robot AI models like SmolVLA and Pi0.
OpenBMB's MiniCPM-V 4.6 packs a 1.3B vision-language model into your phone, outperforming larger rivals with 50% fewer visual encoding FLOPs