Considering Perspectives for Automated Driving Ethics: Collective Risk in Vehicular Motion Planning
Authors: Leon Tolksdorf, Arturo Tejada, Christian Birkner, Nathan van de Wouw
Organizations: Department of Dynamics and Control, Eindhoven University of Technology, Eindhoven, The Netherlands · CARISSMA Institute of Safety in Future Mobility, Technische Hochschule Ingolstadt, Ingolstadt, Germany · TNO, Integrated Vehicle Safety, Helmond, The Netherlands
Recent automated vehicle (AV) motion planning strategies evolve around minimizing risk in road traffic. However, they exclusively consider risk from the AV's perspective and, as such, do not address the ethicality of its decisions for other road users. We argue that this does not reduce the risk of each road user, as risk may be different from the perspective of each road user. Indeed, minimizing the risk from the AV's perspective may not imply that the risk from the perspective of other road users is also being minimized; in fact, it may even increase. To test this hypothesis, we propose an AV motion planning strategy that supports switching risk minimization strategies between all road user perspectives. We find that the risk from the perspective of other road users can generally be considered different to the risk from the AV's perspective. Taking a collective risk perspective, i.e., balancing the risks of all road users, we observe an AV that minimizes overall traffic risk the best, while putting itself at slightly higher risk for the benefit of others, which is consistent with human driving behavior. In addition, adopting a collective risk minimization strategy can also be beneficial to the AV's travel efficiency by acting assertively when other road users maintain a low risk estimate of the AV. Yet, the AV drives conservatively when its planned actions are less predictable to other road users, i.e., associated with high risk. We argue that such behavior is a form of self-reflection and a natural prerequisite for socially acceptable AV behavior. We conclude that to facilitate ethicality in road traffic that includes AVs, the risk-perspective of each road user must be considered in the decision-making of AVs.
Figures & tables
Figure 1: (a) Schematic of a typical traffic scene featuring asymmetric risk. First , the aspect of asymmetric uncertainty is depicted by the colored areas. Upper subfigure: the blue vehicle has the white vehicle well in its sight (denoted by its white/grey perception cone); however, it has limited perception (magenta colored cone) of the cyclist, as the cyclist approaches the blue vehicle from its side. Vice-versa in the lower subfigure: the white vehicle has the cyclist well in its forward sight; however, the blue vehicle is in its blind-spot. The cyclist’s perception is not depicted, though, clearly, it has both vehicles well in its sight. Second , the asymmetric severity can also be deducted. Here, the blue vehicle could collide with its front into the white vehicle’s driver-side door. Such a collision is typically associated with more harm to the white vehicle’s passengers than to the passengers of the blue vehicle. Likewise, if the cyclist collides with the blue vehicle, the consequences of the collision would be much more severe for the cyclist than for the blue vehicle’s passengers. (b) Example for the assumptions: In the upper subfigure, the ego predicts the uncertain trajectories of two objects, displayed for two different time instances. In the lower subfigure, the ego is estimating the uncertainty both objects have about the planned trajectory of the ego vehicle. The wider distributions for object 1 indicate that the ego assess that object 1 is more uncertain about the ego’s trajectory than object 2 . (c) Simulation setup and recording of evaluation metrics for a given scenario with an egoistic risk perspective. The user sets the selector, and a scenario is simulated with that selector setting. While only one risk cost is selected to generate ego behavior, all other risk costs are still recorded (see also Figure 4 ). Note that the ego’s behavior is unaffected by the objects’ risks exclusively for the egoistic risk perspective.
Figure 2: Examples for different risk perspectives (upper subfigures, scenario ZAM_Tjunction-1_25_T-1 ) and uncertainty settings given a collective risk perspective (lower subfigures, scenario ZAM_Zip-1_49_T-1 ). The blue vehicle represents the ego vehicle. The fading colors of vehicles denote a time difference of 0.3s in ZAM_Tjunction and 0.2s in ZAM_Zip . The predictions for a prediction horizon of 2s at the presented time step are highlighted by the yellow-to-red ellipses, where a darker color denotes a higher probability. Lastly, the black trajectories of each object denote the respective true trajectory for the prediction horizon, and the blue trajectory shows the planned ego trajectory for the prediction horizon.
Figure 3: (a) Collective average risk cost for each risk mode (normalized to different scales for visibility). (b) Differences in risk cost allocation between the collective and egoistic perspectives. Negative values indicate a reduction. (c) Histograms of average accumulated collective risk cost at the last time step for ZAM_Tjunction , where, upper row: low uncertainty, center row: moderate uncertainty, bottom row: high uncertainty. Each plot is scaled to contain 18 bins. Note that as we report the average accumulated collective risk cost for each risk-perspective, we obtain nine cases since the uncertainty level always affects the collective risk cost.
Figure 4: Average accumulated risk costs across scenario clusters, risk perspectives, and uncertainty levels. (a) Average accumulated risk costs over all simulation time steps k for the scenario ZAM_Tjunction . (b) Average accumulated risk costs for the final time step k=NT for the scenarios ZAM_Zip (left) and USA_US101 (right).
Risk perspective
Uncertainty Level
Avg. Max. Ref. Error
Avg. Acc. Ref. Error
Avg. Travel. Dist.
Avg. Dist. to Closest Obj.
ZAM_Zip
Egoistic
-
7.56
416.65
92.55
9.31
Collective
low
7.21
406.80
94.16
9.12
mod.
7.33
414.92
93.75
9.44
high
7.80
427.57
91.60
9.52
Altruistic
low
7.39
379.45
99.69
7.81
Table 1: Ego behavior for each scenario cluster, an risk setting. Best values are highlighted in bold font.
Figure 5: An ego vehicle and object o covering of three circles each. As an example, we depict three severity functions sj,l for a collision between the front ego circle j=1 with each object circle l . We omit time indexing for clarity.
Figure 6: Scenario clusters. Top: ZAM_Zip , bottom left: USA_US101 , bottom right: ZAM_Tjunction . The blue depicted vehicle represents the ego.
ZAM_Zip
ZAM_Tjunction
USA_US101
No. of scenarios
71
547
65
Varying road geometry
✗
✗
✓
Varying no. of obj.
✗
✗
✓
Varying init. cond. of obj.
✓
✓
✓
Reference velocity [m/s]
15
10
10
Table 2: Specifics of CommonRoad scenario clusters.
Uncertainty Level
Egoistic Risk Persp. [×103]
Altruistic Risk Persp. [×103]
Collective Risk Persp. [×103]
ZAM_Zip
Low
4.60
15.21
6.69
Moderate
7.25
12.17
8.99
High
11.34
12.87
9.68
ZAM_Tjunction
Low
5.00
3.83
2.50
Supplementary Table S1: Standard deviation of accumulated collective risk cost per scenario cluster. Lowest values per scenario cluster and uncertainty setting are highlighted in bold font.
Autonomous driving develops vehicles that perceive road scenes, predict interactions, and plan and control safe movement with limited or no human input.
In mixed-traffic environments where autonomous and human-driven vehicles may co-exist, motion planning for autonomous vehicles requires anticipating the future behaviors of surrounding human drivers. Existing reinforcement learning-based methods generally directly incorporate the predicted human intents into the observation to enable a proactive planning. However, human intent is inherently uncertain due to the behavioral diversity, perception noise, and partial observability. Treating predicted intends as deterministic states can result in unsafe decisions for autonomous vehicles. To address this problem, we propose Uncertainty-Aware Motion Planning (UAMP), which incorporates uncertainty in human intent prediction for AV decision-making. Specifically, UAMP first introduces a proximity-aware uncertainty estimator to quantify the interaction-conditioned intent uncertainty and constructs an uncertainty-guided joint intent distribution over surrounding human-driven vehicles. Within this uncertainty set, UAMP further introduces Uncertainty-Calibrated Value Learning (UCVL) to correct value function learning biases arising from directly incorporating uncertain human intent predictions into the observation. Extensive experiments in various mixed-traffic scenarios show that UAMP significantly improves safety and driving comfort, while maintaining traffic efficiency compared with existing approaches. The code is released at https://anonymous.4open.science/r/UAMP-5638.
Ming Cheng, Hao Chen, Ziyi Yang +2
Central South University · City University of Macau
Safe motion planning in automated driving requires anticipating evolving traffic risks and deciding when to revise the current planned trajectory. We introduce RiskWorld, a risk-aware world modeling framework for shared occupancy forecasting and selective trajectory replacement. Spatial risk fields and temporal actor context are fused with visual bird's-eye-view features. Flow-guided evolution transports occupancy and scene features, while signed residuals correct occupancy after transport. One forecast is generated per planning step and reused across candidates. Each candidate is compared with a current-state persistence reference, yielding a nonnegative collision-score correction. The trajectory selected by current-world evaluation serves as the planning anchor and is replaced only when additional predicted risk triggers intervention and an alternative satisfies component-wise constraints on predicted risk and trajectory error. Candidate geometries remain unchanged. We evaluate RiskWorld for open-loop planning on nuScenes using camera features, annotation-derived current and historical actor states, and dataset-provided map context. RiskWorld achieves the lowest collision rate at a long evaluation horizon of 3 s, and the second-best average L2 error among various state-of-the-art baselines, while running at 11.5 FPS on a single NVIDIA RTX 4090 with 90.81 M parameters. Within-setting ablations show that RiskWorld achieves lower collision rates than the current-state rescoring baseline, while forecast reuse enables additional candidates to be evaluated at low marginal computational cost.
Rongxiang Zeng, Linsen Cai, Jiafu Zhang +7
RWTH Aachen University, Aachen, Germany · Monash University, Melbourne, Australia · Technical University of Munich (TUM), Munich, Germany
End-to-end autonomous driving resides not in the integration of perception and planning, but rather in the dynamic multi-agent game within a unified representation space. Most existing end-to-end models treat all agents equally, hindering the decoupling of real collision threats from complex backgrounds. To address this issue, We introduce the concept of Risk-Prioritized Game Planning, and propose GameAD, a novel framework that models end-to-end autonomous driving as a risk-aware game problem. The GameAD integrates Risk-Aware Topology Anchoring, Strategic Payload Adapter, Minimax Risk-Aware Sparse Attention, and Risk Consistent Equilibrium Stabilization to enable game theoretic decision making with risk prioritized interactions. We also present the Planning Risk Exposure metric, which quantifies the cumulative risk intensity of planned trajectories over a long horizon for safe autonomous driving. Extensive experiments on the nuScenes and Bench2Drive datasets show that our approach significantly outperforms state-of-the-art methods, especially in terms of trajectory safety.
Kang Ding, Hongsong Wang, Jie Gui +1
School of Cyberspace Security, Southeast University · School of Computer Science and Engineering, Southeast University · Vehicle and Mobility School, Tsinghua University