AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

Reliable Policy Transfer for Safety-Aware End-to-End Driving with Deep Reinforcement Learning

Uddin Md. Borhan, Arif Raza, Zhiliang Lin, Lu Wang, Jianqiang Li, Jie Chen

CVPR 2026 · Reinforcement learning › Visual and representation RL · approach: Scalable RL systems and environments

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
add uncertainty-weighted relational attention (ablation)full framework · CARLA Town10HDcross-track error: 0.76 -> 0.65Improved
add critic ensemble for epistemic uncertainty (ablation)full framework · CARLA Town10HDcollisions per km: 0.009 -> 0.006Improved
add uncertainty-gated entropy (ablation)full framework · CARLA Town10HDcollisions per km: 0.008 -> 0.006Improved
use differentiable multi-objective reward instead of event-driven penalties (ablation)full framework · CARLA Town02driving score: 195.9 -> 214.3Improved
add causal-uncertainty transfer objective (ablation)full framework · CARLA Town02driving score: 194.1 -> 214.3Improved
add MAML meta-initialization (ablation)full framework · CARLA Town02driving score: 200.8 -> 214.3Improved
remove transfer objective or MAML (ablation)full framework · CARLA Town10HDcross-track error: 0.65 -> 0.65No clear effect
shift reward weights toward progress (wp 0.5) (ablation)reward shaping · CARLAroute completion / collisions: +2.1% completion, +0.002/km collisionsMixed
vary entropy base beta0 between 0.5 and 1.0 (ablation)entropy gating · CARLAcollision rate / CTE: collisions <=0.001/km, CTE 0.03No clear effect
use ego-relational state instead of global tensor encodersE2E driving policy · CARLA Town10HDoff-road (%): 10.8 -> 4.1 vs ST-P3Improved
transfer with aligned actions, attention and uncertainty versus target-only trainingTown02 · CARLA Town02success rate (%): 72.1 -> 85.0Improved

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