Author

Machine Heart

Professional AI media and industry service platform

1.0k

Articles

Likes

5.3k

Views

Comments

Latest from Machine Heart

100 recent articles max

Machine Heart

Sep 11, 2026 · Artificial Intelligence

PhyFilter: Physics-Based Feedback Lets Robots Generalize Without Massive Data

Researchers from Beihang University and NTU propose PhyFilter, a plug-and-play physics-informed filter that corrects neural network errors using real-time robot state feedback and known differential structures, enabling zero-shot generalization across quadrupeds, drones, aerial manipulators, and acceleration estimation with only flat-ground simulation training.

PhyFilteraerial manipulationdrone control

0 likes · 18 min read

PhyFilter: Physics-Based Feedback Lets Robots Generalize Without Massive Data

Machine Heart

Sep 11, 2026 · Artificial Intelligence

PhyAgentOS v1.0.0: Executable, Verifiable, Evolvable Harness for Physical Agents

PhyAgentOS v1.0.0 introduces an open-source harness that unifies heterogeneous robots, verifies task outcomes with evidence, enables bounded recovery from failures, and evolves skills through verified experience, achieving 98.6% success on LIBERO and measurable gains on CALVIN and RoboCasa benchmarks.

PhyAgentOSVLAbenchmarks

0 likes · 17 min read

PhyAgentOS v1.0.0: Executable, Verifiable, Evolvable Harness for Physical Agents

Machine Heart

Sep 10, 2026 · Artificial Intelligence

DiffuTester: Accelerating Diffusion LLM Unit Test Generation via Structural Pattern Mining

DiffuTester is a training-free framework that accelerates diffusion language model unit test generation by mining shared structural patterns across test cases via AST, enabling more tokens per denoising step and achieving 2–3× speedup while maintaining coverage across Python, C++, and Java on DiffuCoder and Dream models.

ASTDiffusion LLMEMNLP 2026

0 likes · 9 min read

DiffuTester: Accelerating Diffusion LLM Unit Test Generation via Structural Pattern Mining

Machine Heart

Sep 10, 2026 · Artificial Intelligence

AgentVLN: VLM-as-Brain Architecture for Agentic Robot Navigation at ECCV 2026

AgentVLN introduces a VLM-as-Brain architecture for vision-language navigation, using cross-space representation mapping to translate 3D paths into 2D visual prompts, context-driven self-correction for error recovery, and query-driven perceptual chain-of-thought for active information gathering, achieving state-of-the-art results on R2R-CE and RxR-CE benchmarks with a 3B model deployable on Jetson edge devices.

AgentVLNCross-Space Representation MappingECCV 2026

0 likes · 10 min read

AgentVLN: VLM-as-Brain Architecture for Agentic Robot Navigation at ECCV 2026

Machine Heart

Sep 9, 2026 · Artificial Intelligence

Action Map Policy: Pixel Classification for High-Precision Robot Manipulation

Action Map Policy (AMP) reframes robot manipulation as pixel-level classification by projecting 3D keypoint trajectories onto 2D image planes, enabling cross-entropy loss to model multi-modal action distributions with sub-millimeter precision and single-forward-pass inference.

Action Map PolicyClosed-Loop ControlCross-Entropy

0 likes · 12 min read

Action Map Policy: Pixel Classification for High-Precision Robot Manipulation

Machine Heart

Sep 8, 2026 · Artificial Intelligence

MetaRSI-v1: Meta-Recursive Self-Improvement Unifies Model, Data, Harness RSI

MetaRSI-v1 introduces a unified meta-recursive self-improvement architecture that applies recursive self-improvement to the improvement process itself, unifying Model-RSI, Data-RSI, and Harness-RSI via a Loop Kernel with horizontal and vertical orchestration, four coordinating agents, and five laws, demonstrating 10.9-point average gains on a 3B model and 7.3-point gains on frontier models like GPT-5.6 and Claude Opus 5.

AI architectureData-RSIHarness-RSI

0 likes · 14 min read

MetaRSI-v1: Meta-Recursive Self-Improvement Unifies Model, Data, Harness RSI

Machine Heart

Sep 8, 2026 · Artificial Intelligence

WorkSwarm's Persistent Sessions: Keeping AI Agents Accurate Over 200+ Turns

WorkSwarm's Persistent Session enables AI agents to maintain context, responsibilities, and decisions across hundreds of interaction turns, demonstrated via a 6-hour, 189-turn multi-user Feishu collaboration resolving 8 cross-responsibility conflicts and a 200-turn coding task where the persistent session completed all tasks while the control group failed at turn 156 due to context compression drift.

AI agentsContext DriftContext Management

0 likes · 15 min read

WorkSwarm's Persistent Sessions: Keeping AI Agents Accurate Over 200+ Turns

Machine Heart

Sep 8, 2026 · Artificial Intelligence

Behavior Consistency Beats State Consistency in Text World Models for Agents

The paper introduces BehR, a behavior consistency reward for training text-based world models, showing that optimizing for agent decision alignment rather than text fidelity improves trajectory-level consistency across 16 configurations, reduces false positives in offline evaluation from 42.5% to 9.5%, and enhances lookahead planning for weaker agents.

Behavior ConsistencyEMNLP 2026GRPO

0 likes · 11 min read

Behavior Consistency Beats State Consistency in Text World Models for Agents

Machine Heart

Sep 8, 2026 · Artificial Intelligence

SMELT: Looped Transformers Outperform Baselines Under Fair Budget Matching

The SMELT framework from Tsinghua and ByteDance Seed fairly compares Looped Transformers against baselines by matching compute, parameters, and KV cache, finding that looping the middle 50% of layers twice with a larger depth-width ratio consistently reduces validation loss across scales, saves 6.8–18% training compute, and yields downstream gains beyond loss reduction.

Budget MatchingLLM ArchitectureLooped Transformer

0 likes · 12 min read

SMELT: Looped Transformers Outperform Baselines Under Fair Budget Matching

Machine Heart

Sep 7, 2026 · Artificial Intelligence

LLaDA-Image: Full Diffusion Model Leads Open-Source Text-to-Image via Image-Only Pretraining

LLaDA-Image is a 6B diffusion transformer that learns visual priors from images alone — 90% of its 220M training samples use image-only supervision — then aligns language in later fine-tuning, achieving top open-source scores on Qwen-Image-Bench, supporting both text-to-image and instruction-based editing, and distilling to 2–4 steps via TwinFlow.

Diffusion TransformerImage EditingImage-Only Pretraining

0 likes · 21 min read

LLaDA-Image: Full Diffusion Model Leads Open-Source Text-to-Image via Image-Only Pretraining