cs.ROSep 25, 2026

INTERACT: Interactive Planning for Autonomous Driving via Anchor-Conditioned Prediction and Trust-Region Refinement

Authors: Aron Distelzweig, Andreas Look, Faris Janjoš, Steffen Hagedorn, Luigi Palmieri, Joschka Boedecker

Organizations: Robert Bosch GmbH, Germany · University of Freiburg, Germany · Coburg University, Germany

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Driving in dense urban traffic is interactive: whether a merge or an unprotected turn succeeds depends on how surrounding agents respond to the ego vehicle. Conventional planners predict first and plan second and, therefore, cannot account for this dependency. Methods that integrate prediction and planning either train both jointly, which introduces task interference, or keep them separate and are restricted to a predefined set of proposals. We present INTERACT: Interactive Planning for Autonomous Driving via Anchor-Conditioned Prediction and Trust-Region Refinement. Our key insight is that surrounding agents react to the intent a trajectory expresses rather than to its exact realization, so a single reactive prediction stays valid across an entire family of plans. INTERACT therefore decomposes interactive planning into prediction across driving intents and optimization within each intent. We derive a small set of diverse intents, which we call anchors, from map geometry, query a dedicated ego-conditioned prediction model once per anchor, and refine every anchor with the Cross-Entropy Method under a trust-region penalty that keeps the refined plan close enough to its anchor for the conditioned reaction to still apply. Prediction thus remains a separate model, avoiding task interference, while conditioning on anchors preserves the dependency. Because each anchor is refined continuously, the final plan is not restricted to the anchor set, yet INTERACT requires only one predictor query per anchor rather than one per candidate plan, with all anchors processed in parallel. On the nuPlan and interPlan closed-loop benchmarks, INTERACT sets a new state of the art, with the largest gains precisely in the interactive scenarios that motivate the method. The code will be released upon acceptance.

Figures & tables

Fig. 1: Interactive planning beyond a fixed proposal set. Proposal-based planners (top) score a fixed set of candidate trajectories and select the best one; the executed plan can therefore never be better than the best candidate in that set. Interact (bottom) treats each candidate as an anchor that expresses a driving intent, queries an ego-conditioned predictor once per anchor, and refines the anchor continuously under a trust-region penalty that keeps the reaction valid. Scoring and selection then operate on refined plans, so the final trajectory is no longer confined to the initial set, at the cost of one predictor query per anchor rather than one per candidate plan.
Fig. 2: Overview of INTERACT. 1 Anchor Generation: We generate a diverse set of anchors, representing different ego intents. 2 Anchor-Conditioned Prediction: For each anchor, a conditional prediction model predicts how surrounding agents react to the corresponding ego intent. 3 Trajectory Refinement: Each anchor is independently refined with Cross-Entropy Maximization (CEM) within a local region, preserving its maneuver intent and thereby the validity of the corresponding anchor-conditioned prediction.
Fig. 3: Comparison of our continuous score components (left) with their discrete PDM counterparts (right) on a nuPlan scenario. For every map location, the progress, TTC, and collision terms are evaluated. Color encodes the score value. The discrete terms are piecewise constant: gradients exist only at edges, leaving the search without guidance on the flat plateaus, backing up scores identically to standing still, and every colliding pose is equally bad. The continuous terms provide an informative gradient throughout: progress degrades smoothly behind the ego and pulls laterally toward the goal lane, TTC decays with the closing distance to slower vehicles, and the collision term grades severity by footprint overlap, pointing out of the conflict.
PredictionsMethodVal14 (R) ↑\uparrowTest14-hard (R) ↑\uparrowTest14-random (R) ↑\uparrowDynamic-10 (R) ↑\uparrowinterPlan (R) ↑\uparrowinterPlan-LC (R) ↑\uparrowinterPlan-full (R) ↑\uparrow
NoneUrban Driver [ 24 ]64.1149.9567.1565.984.009.65—
GC-PGP [ 25 ]63.8239.3651.39—14.5534.99—
PlanCNN [ 26 ]72.0052.2067.50——58.35—
PlanTF [ 27 ]76.9561.6179.58—30.5355.18—
PDM-Open [ 4 ]54.2435.8357.23——23.35—
DiffusionPlanner [ 28 ]82.8069.2282.93—25.7625.96—
TABLE I: Comparison of planning performance on the nuPlan benchmarks, grouped by the type of agent prediction the planner consumes. We report closed-loop reactive (R) scores for the Val14, Test14-hard, Test14-random, and Interactive splits, as well as for interPlan, interPlan-LC, and interPlan-full. Higher values indicate better performance. Best and second-best values per column are marked in bold and underlined , respectively. Our method is highlighted . *BIBeR-CV filters candidate plans based on constant-velocity predictions but uses reactive learned predictions afterwards.
Fig. 4: Sensitivity to the trust-region weight λ\lambda on Test14-hard (left) and interPlan-LC (right): closed-loop score (solid), TTC compliance (dashed) and progress along the route (dash-dotted). A small λ\lambda lets the CEM trade safety for progress, a large λ\lambda keeps it close to the anchors and sacrifices progress without further safety gains.
ConfigurationAt- FaultTTCDACMaking ProgressProgress along RouteComfortScore
Test14-hard
Ours97.4389.7198.9096.3285.1081.2585.35
w/o anchors−- 4.25−- 10.00−- 4.43−- 5.92−- 4.17+12.48−- 13.35
w/o refinement+0.74+2.94−- 0.37−- 1.47−- 7.40−- 2.57−- 2.92
w/o continuous score−- 1.47−- 3.310.00+0.37−- 1.16+0.74−- 2.19
interPlan-LC
TABLE II: Component ablation on Test14-hard and the lane-change scenarios of interPlan (interPlan-LC). Each row removes one component of the full approach and reports the difference to the full approach (blue: better, red: worse).
PredictionAt- FaultTTCDACMaking ProgressProgress along RouteComfortScore
Test14-hard
Conditioned (Ours)97.4389.7198.9096.3285.1081.2585.35
Unconditioned+0.37+0.740.00−- 0.370.00+2.21+0.14
Constant velocity−- 0.37−- 1.470.000.00−- 0.19−- 1.47−- 1.24
interPlan-LC
Conditioned (Ours)100.00100.00100.00100.0080.3476.6789.07
TABLE III: Effect of prediction used on Test14-hard and lane-change scenarios of interPlan (interPlan-LC). Rows below ours report the difference to our anchor-conditioned SMART prediction (blue: better, red: worse).
Fig. 5: Sensitivity to the number of anchors NN on Test14-hard (left) and interPlan-LC (right): closed-loop score (solid), at-fault collisions (dashed) and progress along the route (dash-dotted). Going from one to five anchors unlocks progress additional anchors only add risk (at-fault collisions on Test14-hard) without improving the score.
StageRuntime (ms)
Anchor generation32.6
SMART inference17.2
Refinement31.8
TABLE IV: Runtime (ms) per anchor averaged over ten interactive scenarios.
More in this topic

Autonomous Driving Planning

Autonomous driving planning selects safe, feasible vehicle actions and trajectories amid traffic, uncertainty, road rules, and changing goals.

See all 385 papers

Explore similar work

CardsList
Dec 3, 2025

Driving is a Game: Combining Planning and Prediction with Bayesian Iterative Best Response

Autonomous driving planning systems perform nearly perfectly in routine scenarios using lightweight, rule-based methods but still struggle in dense urban traffic, where lane changes and merges require anticipating and influencing other agents. Modern motion predictors offer highly accurate forecasts, yet their integration into planning is mostly rudimental: discarding unsafe plans. Similarly, end-to-end models offer a one-way integration that avoids the challenges of joint prediction and planning modeling under uncertainty. In contrast, game-theoretic formulations offer a principled alternative but have seen limited adoption in autonomous driving. We present Bayesian Iterative Best Response (BIBeR), a framework that unifies motion prediction and game-theoretic planning into a single interaction-aware process. BIBeR is the first to integrate a state-of-the-art predictor into an Iterative Best Response (IBR) loop, repeatedly refining the strategies of the ego vehicle and surrounding agents. This repeated best-response process approximates a Nash equilibrium, enabling bidirectional adaptation where the ego both reacts to and shapes the behavior of others. In addition, our proposed Bayesian confidence estimation quantifies prediction reliability and modulates update strength, more conservative under low confidence and more decisive under high confidence. BIBeR is compatible with modern predictors and planners, combining the transparency of structured planning with the flexibility of learned models. Experiments show that BIBeR achieves an 11% improvement over state-of-the-art planners on highly interactive interPlan lane-change scenarios, while also outperforming existing approaches on standard nuPlan benchmarks.

Aron Distelzweig, Yiwei Wang, Faris Janjoš +5

University of Freiburg, Germany · Bosch Center for Artificial Intelligence, Germany · Technical University of Munich, Germany +2

Autonomous Driving PlanningMulti-Agent Planning
AskPDF
Aug 6, 2026

A Unified Framework for Trajectory Prediction with Explicit Planning and Reaction Decomposition

Trajectory prediction has shifted toward structured formulations with explicit social modeling. However, existing methods inadequately distinguish the functional roles of social influence in trajectory planning. Observing that agents typically form motion plans by anticipating others' future behaviors before making local reactive adjustments, we identify social interactions as playing staged roles, namely planning precedes reaction. We propose INTraJ, a unified framework that decomposes social influence into two stages: a planning stage constructs reference trajectories using future social information, and a reaction stage recovers local adjustments from the residual between full-context prediction and the reference. INTraJ supports both multi-target and single-target paradigms. Extensive experiments on four standard benchmarks, including Argoverse 2, Argoverse 2-ped, ETH/UCY, and SDD, demonstrate consistent improvements, particularly in FDE and long-horizon consistency, with state-of-the-art performance achieved in several settings. INTraJ reframes trajectory prediction as a planning-driven two-stage process, validating that staged social modeling is critical for stable predictions. The code is publicly available at https://github.com/11isnotavailable/INTraJ.

Jiaheng Chen, Jiaxing Li, Tinghe Zhang +1

Software College, Northeastern University Shenyang, China

Pedestrian Trajectory PredictionMulti-Agent Trajectory PredictionAgent Trajectory AnalysisTrajectory PredictionVehicle Trajectory Prediction
AskPDF
Jun 2, 2026

Bridging Predictive Uncertainty and Safe Action: Sample-Conditioned Differentiable Planning for Autonomous Driving

Complex, dynamic, and interactive driving environments pose significant challenges for autonomous driving, primarily due to the pervasive uncertainty of surrounding traffic. A fundamental bottleneck in current systems is the disconnect between highly expressive uncertainty modeling and interpretable, safe motion planning. In this paper, we propose a novel sample-conditioned differentiable planning framework that bridges this gap by explicitly incorporating diffusion-generated future trajectories into the optimization process. Rather than compressing predictions into a single deterministic future or relying on black-box end-to-end architectures, our approach leverages a conditional diffusion model to generate a diverse set of plausible future scenarios. Crucially, these samples are directly fed into a differentiable planner, which explicitly mitigates predictive uncertainty via an empirical Conditional Value-at-Risk (CVaR) tail-risk constraint. This allows the planner to optimize a physically interpretable trajectory that is robust to rare yet safety-critical interactions. Furthermore, we introduce a directed graph representation for scene context that yields substantial improvements in both predictive effectiveness and computational efficiency. Validated through extensive open-loop and closed-loop evaluations on the Waymo Open Motion and Argoverse 2 datasets, our framework significantly outperforms state-of-the-art baselines in safety, efficiency, and ride comfort.

Chengzhen Meng, Pei Liu, Zhiyu Huang +2

Robotics and Autonomous Systems Thrust, The Hong Kong University of Science and Technology, Guangzhou, China, and also the Department of Electronic and Computer Engineering, The Hong Kong University of Science and Technology, Hong Kong SAR, China · Department of Civil and Environmental Engineering, University of California, Los Angeles, CA 90095 USA · School of Mechanical and Aerospace Engineering, Nanyang Technological University, Singapore 639798

Diffusion-Based PlanningRisk-Sensitive PlanningSafe Motion PlanningAutonomous Driving PlanningDecision-Making under Uncertainty
AskPDF