# ALFWorld

Decision models are usually judged by how well they act, yet the same models are pulled toward an easier subgoal when acting on ALFWorld. Used instead as a teacher that sees the whole trajectory after an episode and is asked whether each action helped, JEV raises a Qwen2.5-1.5B policy trained with GRPO from 72.8% to 82.9% success on ALFWorld and from 56.8% to 71.4% on WebShop, while JEV as an actor completes only 42.9% of ALFWorld tasks.

Decision models are usually judged by how well they act, yet the same models are pulled toward an easier subgoal when acting on ALFWorld. Used instead as a teacher that sees the whole trajectory after an episode and is asked whether each action helped, JEV raises a Qwen2.5-1.5B policy trained with GRPO from 72.8% to 82.9% success on ALFWorld and from 56.8% to 71.4% on WebShop, while JEV as an actor completes only 42.9% of ALFWorld tasks.

Long-horizon language-model agent training now builds action preferences inside existing rollout groups, matching on pre-action state plus task progress, and raises mean WebShop success by 3.96 and 5.16 points on Qwen3-4B and 8B.

Long-horizon language-model agent training now builds action preferences inside existing rollout groups, matching on pre-action state plus task progress, and raises mean WebShop success by 3.96 and 5.16 points on Qwen3-4B and 8B.

CB-Agent extracts environment-action decisions from expert trajectories and fits a reusable concept vector to score candidate actions, reaching the highest cross-benchmark average among the evaluated methods at all three Qwen3 sizes on TextWorld, ALFWorld, and ScienceWorld.

CB-Agent extracts environment-action decisions from expert trajectories and fits a reusable concept vector to score candidate actions, reaching the highest cross-benchmark average among the evaluated methods at all three Qwen3 sizes on TextWorld, ALFWorld, and ScienceWorld.

Multi-turn agent distillation now grounds teacher supervision in the interaction state, keeping only executed actions and observations, and the Qwen3-4B student surpasses its 32B teacher on ALFWorld Seen at 43.39%.

Multi-turn agent distillation now grounds teacher supervision in the interaction state, keeping only executed actions and observations, and the Qwen3-4B student surpasses its 32B teacher on ALFWorld Seen at 43.39%.

Swapping agent action selection for a fast decision model like Jev cut per-decision cost by 66.1% to 78.4% and decision time by 73.9% to 81.5%, yet task success held near baseline on ALFWorld and fell more sharply in two mobile agents, so local savings do not automatically become delivered outcomes.

Swapping agent action selection for a fast decision model like Jev cut per-decision cost by 66.1% to 78.4% and decision time by 73.9% to 81.5%, yet task success held near baseline on ALFWorld and fell more sharply in two mobile agents, so local savings do not automatically become delivered outcomes.

RELACE is a critic-free framework that uses a frozen behavior policy to teacher-force score each executed action under both its original context and an outcome-augmented context, normalizes and clips the ratio of these length-normalized scores within each trajectory to reweight discounted returns, and then builds local advantages by comparing weighted returns among actions from matched states, combining them with trajectory-level GRPO supervision; on ALFWorld and WebShop with Qwen2.5-1.5B-Instruct and Qwen2.5-7B-Instruct it reports the highest mean aggregate metrics among the compared methods, including GRPO, GiGPO, HCAPO, and GraphGPO.

RELACE is a critic-free framework that uses a frozen behavior policy to teacher-force score each executed action under both its original context and an outcome-augmented context, normalizes and clips the ratio of these length-normalized scores within each trajectory to reweight discounted returns, and then builds local advantages by comparing weighted returns among actions from matched states, combining them with trajectory-level GRPO supervision; on ALFWorld and WebShop with Qwen2.5-1.5B-Instruct and Qwen2.5-7B-Instruct it reports the highest mean aggregate metrics among the compared methods, including GRPO, GiGPO, HCAPO, and GraphGPO.

UniSkill lets a shared policy act as both the actor that executes environment actions and the proposer that emits skillbank edits (Add, Update, No Edit), and trains the proposer with contrastive action feedback that measures how replacing the retrieved skill changes the current actor's action log-likelihood gap between previously collected successful and failed trajectories, avoiding extra rollouts per proposal, with skill-edit support regularization preserving exploration; it reaches 98.4% success on ALFWorld and 84.7% success on WebShop while sustaining stable joint training over 250 steps.

UniSkill lets a shared policy act as both the actor that executes environment actions and the proposer that emits skillbank edits (Add, Update, No Edit), and trains the proposer with contrastive action feedback that measures how replacing the retrieved skill changes the current actor's action log-likelihood gap between previously collected successful and failed trajectories, avoiding extra rollouts per proposal, with skill-edit support regularization preserving exploration; it reaches 98.4% success on ALFWorld and 84.7% success on WebShop while sustaining stable joint training over 250 steps.