ALFWorld
A widely used evaluation benchmark in embodied agent research. Its tasks involve completing a series of object manipulation operations in a simulated household environment, used to assess agent planning and execution capabilities.
Timeline (last 90 days)
This reproduction study covers four agent task environments—ALFWorld, ScienceWorld, Search-QA, and WebShop—spans three model scales, and compares 10 representative self-distillation methods against a GRPO baseline.
CoPE was validated on five benchmarks: ALFWorld, ScienceWorld, WebShop, τ²-bench, and AppWorld.
SkillScaffold was evaluated on four benchmarks—ALFWorld, WebShop, ScienceWorld, and τ²-Bench—and surpassed strong agentic RL baselines and skill-driven RL baselines on both short-horizon and long-horizon tasks.
On the ID/OOD average success rate on ALFWorld, BranchOPD improves over Vanilla OPD by 13.8 percentage points.
Compared to a scheme with seven independent trajectories, shared prefix reuse reduces student-generated token costs by approximately 19% on ALFWorld.
After targeted training using the ALFWorld and TextCraft benchmarks, the two types of planners performed similarly on plan generation success rate.
The research team validated SAGA on two interactive environment benchmarks, ScienceWorld and ALFWorld, achieving improved task completion performance.
ALFWorld combines text instructions with embodied navigation, requiring agents to perform everyday tasks in a virtual household environment.
On ALFWorld, CTWM achieved consistent token savings under identical paired comparisons.
ALFWorld is a text-based household interaction benchmark used to evaluate language models' planning and tool-use capabilities.
6 more timeline events
All Facts (16)
This reproduction study covers four agent task environments—ALFWorld, ScienceWorld, Search-QA, and WebShop—spans three model scales, and compares 10 representative self-distillation methods against a GRPO baseline.
50%UnverifiedCoPE was validated on five benchmarks: ALFWorld, ScienceWorld, WebShop, τ²-bench, and AppWorld.
50%UnverifiedSkillScaffold was evaluated on four benchmarks—ALFWorld, WebShop, ScienceWorld, and τ²-Bench—and surpassed strong agentic RL baselines and skill-driven RL baselines on both short-horizon and long-horizon tasks.
50%UnverifiedOn the ID/OOD average success rate on ALFWorld, BranchOPD improves over Vanilla OPD by 13.8 percentage points.
50%UnverifiedCompared to a scheme with seven independent trajectories, shared prefix reuse reduces student-generated token costs by approximately 19% on ALFWorld.
50%UnverifiedAfter targeted training using the ALFWorld and TextCraft benchmarks, the two types of planners performed similarly on plan generation success rate.
50%UnverifiedThe research team validated SAGA on two interactive environment benchmarks, ScienceWorld and ALFWorld, achieving improved task completion performance.
50%UnverifiedALFWorld combines text instructions with embodied navigation, requiring agents to perform everyday tasks in a virtual household environment.
50%UnverifiedOn ALFWorld, CTWM achieved consistent token savings under identical paired comparisons.
50%UnverifiedALFWorld is a text-based household interaction benchmark used to evaluate language models' planning and tool-use capabilities.
50%UnverifiedThe research team validated MATE's effectiveness on 134 ALFWorld tasks.
50%UnverifiedMATE's current validation is primarily focused on the ALFWorld benchmark, and its generalization capability in more complex and open-ended real-world environments remains to be further examined.
50%UnverifiedExperiments were conducted in the ALFWorld environment using a frozen 7B model as the backbone.
50%UnverifiedALFWorld is a benchmark environment that combines text adventure games with embodied AI tasks, proposed by Carnegie Mellon University and other institutions in 2021, and is based on the ALFRED dataset.
50%UnverifiedFor the self-evolution of agents, not all components are worth investing optimization resources in; role settings and format rules have almost no marginal contribution to the final success rate in ALFWorld tasks.
50%UnverifiedHARNESSEVO achieved an overall binary success rate of 0.657, compared to 0.642 for both the original harness and flat string evolution.
50%