alfworld
Here are 14 public repositories matching this topic...
Experience memory for LLM agents that act, tested one rung at a time against token-matched placebo controls. Latest published lab result (ALFWORLD-LADDER-R4-V5B, 201 units): mismatched_type vs placebo +1.5 points (95% CI −6.7 to +9.7, equivalent); retrieved_task vs placebo +16.9 points (95% CI +5.3 to +28.5, positive).
-
Updated
Nov 20, 2026 - Python
Using VLM-based visual question answering to perceive scenes and control robots in MuJoCo simulation.
-
Updated
Oct 4, 2026 - Python
Official code & benchmark for ALFWorld-Para and BINDIntent (“Binding Paraphrased Instructions to Encountered Objects in sLLM Agents under Partial Observability”)
-
Updated
Nov 6, 2026 - PDDL
[arXiv] ExpHarness: Model-Agnostic Experience Learning through a Trainable Harness
-
Updated
Oct 4, 2026 - Python
WM-PPO: Learning to act by valuing predicted state transitions. World-model predictions + critic-based valuation for calibrated policy and value learning under sparse terminal rewards. arXiv:2611.12200
-
Updated
Nov 10, 2026 - Python
Execution-time action control for ReAct language agents with stage-aware guards, wrong-object prevention, and action redirection in ALFWorld.
-
Updated
Jun 16, 2026 - Python
Metacognition for LLM Agents adapting at test time to out-of-distribution tasks from limited data
-
Updated
Nov 20, 2026 - Python
Single-GPU ALFWorld agent post-training with Qwen2.5-1.5B: expert trajectory SFT + online GRPO, task success 6.25% to 68.75% to 89.58%, with open data, adapters and reproducible evaluation.
-
Updated
Oct 16, 2026 - Python
Audited Agent-G² reproduction: GRPO Gaussian guidance, ALFWorld, RTX 4090 LoRA evidence, reproducibility gates
-
Updated
Sep 28, 2026 - Python
Multi-turn GRPO on ALFWorld — SIRI+Replay skill internalization. Qwen2.5-1.5B-Instruct + vLLM + LoRA.
-
Updated
Jun 15, 2026 - Python
An LLM-driven agent that explores simulated environments (ALFWorld) to infer inhabitant occupations using DSPy and Chain-of-Thought reasoning.
-
Updated
Nov 23, 2025 - Python
BELT: reusable belief representations for early stopping in LLM reasoning. Code, trained checkpoints, and archived numerical evidence.
-
Updated
Nov 12, 2026 - Python
Add this topic to your repo
To associate your repository with the alfworld topic, visit your repo's landing page and select "manage topics."