AI Paper Atlas

Research map / Reinforcement learning / Imitation learning

GRACE: A Language Model Framework for Explainable Inverse Reinforcement Learning

Silvia Sapora, R Devon Hjelm, Omar Attia, Alexander T Toshev, Bogdan Mazoure

ICLR 2026 · Reinforcement learning › Imitation learning · approach: Reward learning from demonstrations · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use GRACE rewards instead of GAIL with 2000 trajectoriesPPO on BabyAI · BabyAI Multi-tasksuccess rate: 0.31 -> 0.92Improved
use GRACE rewards instead of GAIL with 2000 trajectoriesPPO on BabyAI · BabyAI GoToRedBallsuccess rate: 0.35 -> 1.00Improved
use GRACE reward instead of ground-truth sparse rewardPPO on BabyAI · BabyAI OpenMatchingDoor (new)success rate: 0.79 -> 0.35Hurt
use GRACE reward instead of ground-truth sparse rewardPPO on BabyAI · BabyAI OpenDoorLocsuccess rate: 0.39 -> 1.00Improved
use GRACE reward instead of ground-truth sparse rewardPPO on BabyAI · BabyAI Multi-tasksuccess rate: 0.95 -> 0.92No clear effect
replace GPT-4o with Qwen3-Coder-30B as GRACE LLMGRACE · MuJoCo Antaverage return: 5707 ± 210 -> 6085 ± 804No clear effect
use GRACE rewards instead of GAIL with 10 trajectoriesPPO on MuJoCo · MuJoCo Walkeraverage return: 790 ± 90 -> 2072 ± 576Improved
use GRACE (GPT-4o) instead of AIRL with 200 trajectoriesPPO on MuJoCo · MuJoCo Antaverage return: 4308 ± 306 -> 5707 ± 210Improved
use GRACE (GPT-4o) instead of GAIL with 200 trajectoriesPPO on MuJoCo · MuJoCo Humanoidaverage return: 6521 ± 337 -> 5809 ± 106Hurt
add targeted reward shaping in Phase 3 (ablation)GRACE · BabyAI OpenTwoDoorssuccess rate / learning speedImproved
increase number of expert trajectories to eight (ablation)GRACE reward recovery · BabyAIreward accuracy: perfect scores achieved using only eight expert trajectoriesImproved
add negative trajectories (one negative with enough expert trajectories) (ablation)GRACE reward recovery · BabyAIreward accuracy: accuracy of 0.95 with a single negative trajectoryImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy