AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

Diffused Task-Agnostic Milestone Planner

Mineui Hong, Minjae Kang, Songhwai Oh

NeurIPS 2023 · Reinforcement learning › Visual and representation RL · approach: RL for combinatorial routing · 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
Plan temporally sparse latent milestones and follow them with a goal-conditioned actorDTAMP versus offline RL and sequence-modeling baselines · D4RL Antmazescore: 85.0$\pm$4.84 vs 61.7Improved
Plan latent milestones rather than whole trajectoriesDTAMP versus sequence modeling baselines · Antmaze long-horizon tasksscore: DTAMP 93.3$\pm$0.94; DT, TT, DD 0.0Improved
Use diffusion-generated milestones for varying goalsDTAMP versus multi-goal baselines · Multi-goal Antmaze averagescore: 70.3$\pm$5.86 vs 49.2Improved
Re-plan the remaining milestones after reaching a milestoneDTAMP · CALVIN averagesuccess rate: 54.8$\pm$4.37 -> 62.7$\pm$3.60Improved
Add reconstruction loss when training the encoder (ablation)DTAMP · CALVIN averagesuccess rate: 50.9 -> 54.8Improved
Add goal-conditioned imitation learning loss alongside reconstruction loss (ablation)DTAMP · CALVIN averagesuccess rate: 49.1 -> 54.8Improved
Use skill-based policy (ablation)DTAMP · CALVIN averagesuccess rate: 31.8 -> 54.8Improved
Add reconstruction loss when training the encoder (ablation)DTAMP · CALVIN 1 tasksuccess rate: 86.7 -> 82.4Hurt
Guide diffusion toward smaller temporal intervals between milestones (ablation)DTAMP milestone planner · Simplified open Antmazetimesteps to reach goalImproved
Apply minimum-temporal-distance diffusion guidance (ablation)DTAMP milestone planner · Simplified open Antmazesuccess rateNo clear effect
Train an encoder using both reconstruction and goal-conditioned imitation learning losses (ablation)DTAMP · CALVIN multi-task planningsuccess rate: 86.7 vs 82.4; 43.3 vs 52.4; 17.2 vs 29.7Mixed
Use a feedback actor and plan milestones once at episode startDTAMP versus Diffuser · maze2d-umaze-v1 under increasing stochasticitynormalized score: DTAMP 97.6, 99.0, 99.4; Diffuser 61.9, 29.6, 18.5Improved

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