AI Paper Atlas

Research map / Reinforcement learning / Policy gradient and actor-critic

Diminishing Return of Value Expansion Methods

Daniel Palenicek, Michael Lutter, Joao Carvalho, Daniel Dennert, Faran Ahmad, Jan Peters

TPAMI 2026 · Reinforcement learning › Policy gradient and actor-critic · approach: Temporal-difference and discounting theory · 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
increase rollout horizon beyond 5 stepsSAC critic/actor expansion (oracle dynamics) · Brax continuous control (aggregated)IQM normalized return: peak performance at most H=5; no further increase for more than 5 stepsNo clear effect
use oracle dynamics instead of learned ensembleSAC-CE · Brax continuous control (aggregated)final performance: up to ~20% higher final performanceImproved
use oracle dynamics instead of learned ensembleSAC-AE · Brax continuous control (aggregated)final performance and learning speed: neural network model learns slightly fasterHurt
increase rollout horizon for actor expansionSAC-AE · Brax continuous controlreturn: performance degrades significantly for longer horizonsHurt
replace learned model rollouts with model-free RetraceSAC-CE (learned model) · Brax continuous control (aggregated)sample efficiency / wall-clock: up to 15x faster in wall-clock time at H=5No clear effect
use deterministic policy (DDPG) so target variance does not grow with H (ablation)DDPG-CE / DDPG-AE · Brax continuous controlreturn: same diminishing returns as SACNo clear effect
average Q-targets over particles (30 for CE, 10 for AE) (ablation)SAC-CE and SAC-AE · Brax continuous controlreturn: no qualitative difference in training performanceNo clear effect
vary discount factor gamma (ablation)SAC value expansion · InvertedPendulumreturn: diminishing returns obvious in all experiments regardless of gammaNo clear effect
increase rollout horizon of oracle modelDQN-CE and DQN-Retrace · MinAtar (4 games, aggregated)return: do not benefit from longer rollout horizonsNo clear effect
increase rollout horizon on target distributionSAC H-step target · Brax continuous controlWasserstein distance to true return: distance decreases with horizon; target variance increases by orders of magnitudeImproved
increase horizon on critic gradientsSAC-CE · Brax continuous controlcritic gradient std: std between 10^-3 and 10^-1No clear effect
increase horizon on actor gradientsSAC-AE · HalfCheetahactor gradient std: explodes for horizons >=3 (oracle) and 20 (learned)Hurt

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