Google released ADK Python v2.12.0, updating its Agent Development Kit for building agents in Python. It is a maintained release in Google's agent tooling line.
ChunkAct introduces action chunking for agentic robotics, grouping actions to improve policy efficiency and execution. It targets real-world robotic agent control.
Tev introduces a method for scaling long-horizon agentic training environments using dense progress rewards. The paper targets improved training signal for long-horizon agent tasks.
PixelMoE applies LLM-style mixture-of-experts to pixel-space diffusion, aiming to scale image generation without a VAE latent stage. Posted to Hugging Face Papers.
iFixAi offers independent auditing of AI agents, runnable by humans or the agents themselves, delivering verification within 120 seconds. It gained over 2,100 stars today, reflecting demand for agent verification tooling.