AI Paper Atlas

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

Test-driven Reinforcement Learning in Continuous Control

Zhao Yu, Xiuping Wu, Liangjun Ke

AAAI 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
learn rewards from test functions (TdRL) instead of using oracle handcrafted rewardSAC · DM-Control Walker-Stand, Walker-Run, Cheetah-Run, Quadruped-Runepisode returnNo clear effect
replace SAC with PPO as policy optimizer in TdRLTdRL · DM-Control tasksepisode returnMixed
use pass-fail tests to define per-metric thresholds instead of weighted rewardTdRL · DM-Control Walker-Runmulti-objective satisfactionImproved
trajectory-level test evaluation instead of state-action rewardsTdRL · DM-Control Walker-RunstabilityHurt
add return penalty term L_Penalty (ablation)TdRL · DM-Control (Figure 4a)episode returnImproved
learn trajectory return first then decompose to reward, instead of direct reward learning (ablation)TdRL · DM-Control (Figure 4a)episode returnImproved
use a moderate multiple K_ES in early stop (not too large or small) (ablation)TdRL-ES · DM-Control (Figure 4b)episode returnImproved
TdRL-GN versus TdRL-ES variants of balancing the two losses (ablation)TdRL · DM-Control tasksepisode returnNo 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