Comfort by Construction: Adaptive, Comfort-Bounded Action Spaces for Learned Driving Policies
On the map: End-to-end autonomous driving → Reinforcement learning for driving
Abstract (arXiv)
Data-driven driving simulators command accelerations and steering rates from a fixed grid without constraining the realized accelerations and jerks. As a result, reinforcement-learning policies inflate safety metrics through abrupt, last-second maneuvers that lie far outside the range of human driving and would be unacceptable to occupants of a real vehicle, so the metrics measure simulator permissiveness rather than policy quality. Enforcing comfort bounds naively is not enough: lateral limits shrink quadratically with speed, so clamping a static grid saturates it and destroys fine-grained control ("grid collapse"). We propose an adaptive action parameterization that rediscretizes the grid at every step to span exactly the per-step feasible control set, via closed-form inversion of the lateral-jerk constraint. We further present PufferDrive-Editor, a browser-based tool to audit realized kinematics and author kinematically challenging scenes. On the Waymo Open Motion Dataset and a hand-authored slalom, our adaptive model holds comfort violations below 1% while outperforming clipped-grid and direct-jerk baselines in navigability.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Reinforcement learning for driving, one of 9 directions in End-to-end autonomous driving. Builds on 16 of its 30 references that are in the map. First posted on arXiv 11 Sep 2026.
Builds on
16 of 30 references.
References · 16
- 2017Proximal Policy Optimization Algorithmscited by 15,215
- 2021Large Scale Interactive Motion Forecasting for Autonomous Driving : The Waymo Open Motion Datasetcited by 537
- 2017Safe Reinforcement Learning via Shieldingcited by 449
- 2019End-to-End Safe Reinforcement Learning through Barrier Functions for Safety-Critical Continuous Control Taskscited by 289
- 2018Safe Exploration in Continuous Action Spacescited by 247
- 2023Waymax: An Accelerated, Data-Driven Simulator for Large-Scale Autonomous Driving Researchcited by 160
- 2023The Waymo Open Sim Agents Challengecited by 86
- 2022BITS: Bi-level Imitation for Traffic Simulationcited by 80
- 2025Robust Autonomy Emerges from Self-Playcited by 67
- 2022Nocturne: a scalable driving benchmark for bringing multi-agent learning one step closer to the real worldcited by 56
- 2024GPUDrive: Data-driven, multi-agent driving simulation at 1 million FPScited by 45
- 2024PufferLib: Making Reinforcement Learning Libraries and Environments Play Nicecited by 24
4 more
- 2020Self-Driving like a Human driver instead of a Robocar: Personalized comfortable driving experience for autonomous vehiclescited by 20
- 2024Action Mapping for Reinforcement Learning in Continuous Environments with Constraintscited by 5
- 2024Deep Stochastic Kinematic Models for Probabilistic Motion Forecasting in Trafficcited by 2
- 2024Combining Deep Reinforcement Learning with a Jerk-Bounded Trajectory Generator for Kinematically Constrained Motion Planningcited by 2
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2024ComDrive: Comfort-Oriented End-to-End Autonomous Driving
- 2026Beyond Safe Execution: Safe Online Reinforcement Learning for Autonomous Driving with Value Calibration and Adaptive Guidance
- 2024RACER: Epistemic Risk-Sensitive RL Enables Fast Driving with Fewer Crashes
- 2025Learning to Drive Safely with Hybrid Options
- 2026Dynamics Distillation for Efficient and Transferable Control Learning
- 2021Minimizing Safety Interference for Safe and Comfortable Automated Driving with Distributional Reinforcement Learning
- 2025DriveDPO: Policy Learning via Safety DPO For End-to-End Autonomous Driving
- 2026YOUDrive: Driving Style as a Composable Direction for End-to-End Autonomous Driving