Despite the widespread adoption of autonomous vehicle technologies, reliable navigation within cluttered environments remains arduous due to the necessity of interaction-aware strategies; example scenarios include: narrow passages with multiple agents, lane changing in dense traffic, and crossing unsignalized intersections. These scenarios often require highly proactive strategies, simultaneously demanding interpretability and quick adaptability, which challenge end-to-end and data-driven methods. An effective strategy requires: scalability to multiple agents, adaptability to changes in other agents' behavior, and harmony with surrounding traffic without freezing or becoming overly conservative. To meet these requirements, we present SIGMO, a sigmoid-gated interaction-aware motion planner, with two main contributions. First, we formulate polynomial sigmoid functions that analytically estimate yield scores of surrounding agents using physics-grounded interaction features that capture driving intuitions. Second, we integrate the polynomial sigmoid functions into a receding-horizon planner, which facilitates interaction-aware planning with arbitrary trajectory predictions. We validate SIGMO in interaction-intensive simulations and hardware experiments. The results show that SIGMO balances proactive and defensive behavior while remaining computationally lightweight, serving as a strong baseline in a mobility stack. The simulation and hardware experiment videos are available here: https://sites.google.com/view/sigmoid-polygate.
Figures & tables
Fig. 1: Motivating scenario : navigating a narrow street where parked vehicles (black) force the ego vehicle and an oncoming agent (blue) into a shared corridor (red), while pedestrians move in different directions. No independently safe gap exists; the ego vehicle must determine whether the oncoming agent and pedestrians are likely to yield, proactively claim the corridor when appropriate, and defer otherwise.
Score
Feature
Sign
μ
τ
sTTC
TTC (s)
+
3.0
0.5
sΣa
Σa (m/s 2 )
−
0.0
1.0
sspeed
c (ratio)
−
0.0
1.0
spath
π (m)
+
0.0
3.0
sprox
dbody (body-radii)
+
4.0
0.5
Polygate (combination) weights
TABLE I: Sigmoid score parameters.
Term (Symbol)
Definition
Lateral goal tracking Jy-goal(i)
T1∑t=1T(yi,t−ygoal)2
Lateral velocity damping Jlat-v(i)
T1∑t=1T(vi,tsinθi,t)2
Speed deviation Jspeed(i)
T1∑t=1T(vi,t−vdes)2
Control effort Jctrl(i)
T1∑t=1T(ai,t2+ϕi,t2)
Jerk Jjerk(i)
T1∑t=0T−1(Δai,t)2
Steering rate Jsteer(i)
T1∑t=0T−1(Δϕi,t)2
TABLE II : Nominal Cost Terms
Fig. 2 : Risks with and without polygate in a merge scenario. Interaction-aware risk in ( 21 ) essentially diminishes along with the yield probability of the agent (red) to the ego vehicle (blue).
Fig. 3 : SIGMO trajectory in narrow passage scenario. SIGMO with the polygate estimates the responsiveness of other agents, successfully navigating through a cluttered corridor. In contrast, the absence of polygate purely relies on the constant prediction, stranding the ego vehicle in the cluttered corridor. The color of an agent indicates the estimated risk to the ego vehicle (i.e., red ↑→ risk ↑ ).
Fig. 4 : SIGMO trajectory in merge scenario. SIGMO with the polygate estimates yield probability of the agent in the target lane, successfully making space and merging. In contrast, the pure constant prediction does not solely predict a yield behavior, having the ego vehicle remain in the source lane. The color of an agent indicates the estimated risk to the ego vehicle (i.e., red ↑→ risk ↑ ).
Fig. 5 : Yield probability estimation in merge scenario. Agent A4 is the key agent that the ego vehicle interacts with to make space to merge. SIGMO with the polygate nudges A4 to slow down, and the yield probability increases as A4 decelerates, leading the ego vehicle to decide to merge.
Fig. 6 : Hardware experiments with F1Tenth [ 25 ] . The ego vehicle interacts with the pedestrian to merge from the ramp. The first image is with Naive MPPI and the rest are with SIGMO (showcasing various behaviors). All the experiments are with the same hardware settings (e.g., localization and perception) as well as the pedestrian prediction. Each pair of red circles indicates the footprints of the pedestrian.
Model
Succ.
Coll.
T.O.
Min-d
tˉcomp
Stall [s]
[%]
[%]
[%]
[ m ]
[s]
Narrow passage
MPPI
69
26
5
0.37
71.8
18.1 (47)
SIGMO (ours)
73
25
2
0.36
70.6
13.7 (43)
[1pt/2pt] Oracle-yield (yield est. bound)
70
30
0
0.36
69.0
11.3 (36)
Oracle-traj (performance bound)
93
7
0
0.57
66.5
8.2 (17)
TABLE III : Monte-Carlo evaluation ( N=100 each).
Method
Succ. [%]
Coll. [%]
OOR [%]
T.O. [%]
SIGMO (ours)
92.7
7.3
0.0
0.0
IANN-MPPI [ 15 ]
89.3
10.7
0.0
0.0
TABLE IV : Performance comparison in MetaDrive on 150 evaluation scenarios. OOR denotes Out-of-Road events.
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.
Aron Distelzweig, Andreas Look, Faris Janjoš +3
Robert Bosch GmbH, Germany · University of Freiburg, Germany · Coburg University, Germany
Safe and explainable motion planning remains a central challenge in autonomous driving. While rule-based planners offer predictable and explainable behavior, they often fail to grasp the complexity and uncertainty of real-world traffic. Conversely, learned planners exhibit strong adaptability but suffer from reduced transparency and occasional safety violations. We introduce Mosaic, a framework for structured decision-making that integrates both paradigms through arbitration graphs. By decoupling trajectory verification and selection from the generation of trajectories by individual planners, every decision becomes transparent and traceable. This separation lets verification and trajectory selection contribute independently: centralized verification acts as a safety floor, reducing at-fault collisions from 25 for each standalone planner to 16. In contrast, per-step trajectory selection acts as a performance ceiling, combining the complementary strengths of a rule-based and a learned planner. In experimental evaluation on nuPlan, Mosaic achieves 95.56 CLS-NR and 94.18 CLS-R on the Val14 closed-loop benchmark, setting a new state of the art. On the interPlan benchmark, focused on highly interactive and out-of-distribution scenarios, Mosaic scores 54.10 CLS-R, outperforming its best constituent planner by 22.8% -- all without retraining or requiring additional data. The code is available at github.com/KIT-MRT/mosaic.
Nick Le Large, Marlon Steiner, Lingguang Wang +4
Institute of Measurement and Control Systems, Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany · FZI Research Center for Information Technology, Karlsruhe, Germany
Safe autonomous driving requires more than recognizing the current scene: the ego vehicle must anticipate how surrounding traffic participants will evolve under its own motion. However, existing end-to-end driving models typically encode scenes into dense visual or BEV features, leaving object-level interactions implicit. We propose (INTERACT INTERaction-aware Action-Conditioned Trajectory learning), a three-stage framework that learns a slot-based object-centric world model from unlabeled driving videos and transfers it to trajectory planning. INTERACT first learns recurrent slots that group dense visual features into temporally structured scene-entity representations without object annotations. It then infers latent ego actions from slot context and video transitions, and uses them to condition future slot prediction, capturing how ego motion and surrounding entities co-evolve. Finally, the learned slots, latent actions, and future slot predictions are distilled into a trajectory planner to guide trajectory generation and selection. Extensive experiments on NAVSIM and nuScenes validate the effectiveness of INTERACT. In particular, it achieves state-of-the-art performance on NAVSIM, with improvements in safety-critical collision metrics.
Zhan Chen, Songyan Zhang, Wenhui Huang +5
Nanyang Technological University, Singapore · Tongji University, Shanghai, China