AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

From Language to Goals: Inverse Reinforcement Learning for Vision-Based Instruction Following

Justin Fu, Anoop Korattikara, Sergey Levine, Sergio Guadarrama

ICLR 2019 · Reinforcement learning › Visual and representation RL · approach: Goal-conditioned policy learning · 65 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
Learn a language-conditioned reward and re-plan, rather than directly clone a language-conditioned policyLC-RL versus Optimal Policy Cloning · Test-Housesuccess rate (%): 8.5 -> 36.4Improved
Learn a language-conditioned reward with IRL rather than use the GAIL discriminator as rewardLC-RL versus GAIL-Exact · Test-Housesuccess rate (%): 28.3 -> 36.4Improved
Use oracle ground-truth reward regression instead of IRL reward learningReward Regression versus LC-RL · Test-Housesuccess rate (%): 36.4 -> 65.7Improved
Re-optimize the learned reward with DQN rather than Q-iterationLC-RL · Test-Tasksuccess rate (%): 51.9 -> 7.7Hurt
Add value-function reward shaping (ablation)DQN reward optimization of LC-RL · Test-Tasksuccess rate (%): 7.7 -> 14.1Improved
Use global pooling rather than a fully connected image embedding (ablation)LC-RL architecture · Test-Tasksuccess rate (%): 38.4 -> 49.7Improved
Use global pooling rather than FiLM for the image embedding (ablation)LC-RL architecture · Test-Tasksuccess rate (%): 43.2 -> 49.7Improved
Use multiplication instead of addition to combine language and image embeddings (ablation)Pooling image architecture · Test-Tasksuccess rate (%): 45.5 -> 49.7Mixed
Use multiplication instead of concatenation to combine language and image embeddings (ablation)Pooling image architecture · Test-Tasksuccess rate (%): 48.5 -> 49.7No clear effect
Increase image size from 32 by 24 to 64 by 64 (ablation)LC-RL with Pooling, Mult · Test-Tasksuccess rate (%): 49.7 -> 51.1No clear effect
Increase the number of CNN filters or embedding sizes (ablation)LC-RL · LC-RL environmentperformanceNo clear effect

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