SIGMO: Sigmoid Gated Interaction-Aware Motion Planning

TL;DR SIGMO estimates how likely each nearby car or pedestrian is to yield from five hand-designed sigmoid features, and uses that estimate to scale each agent's risk inside an MPPI planner. In a packed-merge simulation it raises success from 26% to 38% over plain MPPI, and it runs at 15 Hz on a 1/10-scale F1Tenth car.
What they do
In a narrow street or a packed merge there is often no gap that is safe on its own, so the ego vehicle has to negotiate: push in when others will yield, back off when they won’t. SIGMO adds this to a standard sampling-based planner (MPPI) without a learned interaction model. An analytic estimate of each agent’s probability of yielding scales down that agent’s risk in the cost.
How it works
- Polygate. Five sigmoid scores feed the estimate: time-to-collision along the predicted trajectories, the agent’s recent acceleration (braking suggests yielding), closing speed, path priority (who reaches the conflict point first) and an elliptical body-frame proximity gate. Each sigmoid’s inflection point and sharpness has a physical meaning. The scores are fused into one yield probability, which is pulled toward 0.3 for adjacent agents.
- Gated risk. For every MPPI sample, the overlap between Gaussian footprints of the ego and an agent gives a spatial risk. A multiplier that falls as the yield probability rises scales it, but the risk is never removed entirely. A separate one-second collision penalty does not depend on the gate.
- The features come from each candidate rollout’s geometry, so any predictor works (constant velocity here) and no prediction is re-run per sample.
Results
- Narrow passage with pedestrians and cars (100 seeds): 73% success against 69% for MPPI, with 25% against 26% collisions, and stall time down 24.3%. An oracle given the agents’ true cooperativeness reaches 70%.
- Packed merge: 38% success against 26%, both with 2% collisions. Nudging lowers the minimum distance from 0.82 m to 0.62 m. An oracle with true reactive trajectories reaches 69%.
- MetaDrive, 150 scenarios initialized from the Waymo Open Dataset: 92.7% success against 89.3% for IANN-MPPI, which uses a learned Social GAN interaction model.
- F1Tenth car with lidar: 15 Hz with 2000 samples. Merging from a ramp past a pedestrian, naive MPPI fails to merge, while SIGMO merges or surrenders behind the pedestrian. Shown qualitatively only.
Limitations
The simulated agents follow simple behavior models that may not reflect real human behavior. Because SIGMO builds on MPPI, it has no hard collision constraints. The hardware tests are a few qualitative runs on a scale car.
Figures




Abstract
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.
Summaries are written automatically from each paper's text. Figures and videos come from the paper and its project page; all credit goes to the authors.
