AITOPAI热报
ModelsOfficialSelected73°

Google AI releases EnvHarness

Google AI Introduces EnvHarness: A Programmable Layer That Turns Static Agent Environments Into Adaptive Training Worlds

官方一手marktechpost@Asif Razzaq
Blog/Media
Read the original
Why we picked it

EnvHarness improves AI agent training efficiency by automatically adapting environments, potentially reducing development costs for researchers and companies in ASEAN.

Google Cloud AI Research, with Washington University in St. Louis and UNC Chapel Hill, has released EnvHarness, an Apache-2.0 layer that turns static agent benchmarks into adaptive training environments. The tool wraps frozen environments through standard reset()/step() interfaces, leaving tasks and verifiers untouched. Across five benchmarks, mined skills gained up to 9.0 points on held-out tasks with 9.8% fewer execution steps.

Image · marktechpost
Original · marktechpost

Google AI Introduces EnvHarness: A Programmable Layer That Turns Static Agent Environments Into Adaptive Training Worlds

Google Cloud AI Research, with Washington University in St. Louis and UNC Chapel Hill, has released EnvHarness, an Apache-2.0 layer that turns a static agent benchmark into one that adapts to the policy training on it. It wraps a frozen environment through the standard reset()/step() interface, so tasks and human-built verifiers stay untouched — and an LLM designer, EnvRigger, writes those wrappers automatically against flaws diagnosed in the agent's own rollouts. Across five benchmarks, mined skills gain up to 9.0 points on held-out tasks with 9.8% fewer execution steps. The post Google AI Introduces EnvHarness: A Programmable Layer That Turns Static Agent Environments Into Adaptive Training Worlds appeared first on MarkTechPost .

#EnvHarness#Google AI#agent training#benchmark#adaptive environments

Keep reading

Recent stories with the same tag or section
ModelsBuildersOfficialSelected00:32
EnvHarness: Google Research demo on co-evolving agent training environments at COLM 2026

Adaptive environments change how agents are trained: instead of evaluating against fixed benchmarks, the training tasks evolve alongside the agent, which can reduce overfitting to static test sets. Researchers and engineers building agent training pipelines can attend the booth demo or look for the paper to assess whether EnvHarness fits their setup. Comparisons with existing static environment toolchains will likely emerge once the framework is released.

#EnvHarness#Google Research#COLM 2026#agents
官方一手Google Research@GoogleResearchSource
ProductsBuildersOfficialSelected03:13
Google Research presents EnvHarness at COLM 2026

Dynamic environment generation attacks a real bottleneck: agents tuned on static setups overfit and collapse when deployed on varied tasks. Practitioners building RL-trained agents can use EnvHarness's plug-in design to keep training environments shifting, improving generalisation without hand-crafting new scenarios. The COLM booth session offers a direct look at the architecture for researchers and engineers attending the conference.

#Google Research#EnvHarness#agents#reinforcement learning
官方一手Google Research@GoogleResearchSource