Bridging the Discrete-Time Safety Gap

28 concepts28 papers1 proposal

In traditional motion planning, safety is often treated as a checklist of discrete moments. A planner might verify that a robot is safe at $t=1.0$ seconds and again at $t=1.1$ seconds. However, as Jeff Erickson et al. noted in their foundational work on kinetic data structures, this discrete sampling creates a "safety gap." If an object moves fast enough or an obstacle is thin enough, the robot can mathematically "jump" through the obstacle between checks—a phenomenon known as tunneling.

To move beyond this "false optimism," as Alexandre Amice et al. describe it, researchers have developed methods to guarantee safety across the entire continuous duration of a trajectory.

The Geometry of Continuous Safety

One way to bridge the gap is to ensure the robot never leaves a "safe volume" in space-time. Instead of checking points, these methods optimize trajectories within geometric corridors.

  • Safe Flight Corridors: H. Jin Kim et al. utilize the Relative Safe Flight Corridor (RSFC), which linearizes collision constraints. By representing trajectories as Bézier curves, they leverage the convex hull property: if the control points of a Bézier curve lie within a safe convex corridor, the entire continuous path is guaranteed to stay within that corridor.
  • Semantic Corridors: For urban environments, Wenchao Ding et al. introduced the Spatio-temporal Semantic Corridor (SSC). This structure translates traffic rules and moving obstacles into a series of interconnected 3D "cubes" in space-time, ensuring the vehicle follows a continuous safe path rather than just hitting safe waypoints.
  • Flexible Prisms: To improve maneuverability, Srujan Deolasee et al. proposed trapezoidal prism corridors, which offer more volume than standard boxes while maintaining the mathematical guarantees required to prevent tunneling.

Bounding the Unknown

When the environment or the robot's motion is less predictable, researchers use mathematical bounds to "fill in" the intervals between samples.

Luigi Palmieri et al. address the safety gap by incorporating a robot's physical limits—such as maximum velocity and acceleration—directly into the collision check. By calculating the maximum possible displacement between two time steps, the authors can define an "action radius" that ensures the robot cannot physically reach an obstacle between samples.

Similarly, S Roberts et al. use Lipschitz optimization to provide conservative guarantees for stochastic trajectories. Their approach ensures that even with noisy data, the probability of a collision occurring at any point in continuous time remains below a strict threshold. For complex manipulators, Patrick D. Holmes et al. use Reachability-based Trajectory Design (ARMTD) to compute the set of all possible positions a robot arm could occupy, ensuring this entire "reachable set" remains collision-free.

Exact Continuous Collision Detection

Rather than just bounding the motion, some frameworks treat time as a fluid variable to identify the exact moment of impact. This is known as Continuous Collision Detection (CCD).

Yi‐King Choi et al. developed a framework for CCD between objects modeled by quadric surfaces. By using algebraic methods to solve for the first point of contact, they eliminate the geometric errors found in traditional triangle-mesh approximations. In the realm of simulation, Xiaohan Ye et al. demonstrate that CCD is essential for physical accuracy; without it, high-velocity robots in simulators might pass through the very objects they are meant to manipulate.

Why It Works: Differentiability and Optimization

The shift toward continuous safety is not just about robustness; it is about making safety "visible" to optimization algorithms. Traditional discrete checks are often non-differentiable—a robot is either colliding or it isn't.

Modern approaches like CIAO, proposed by Luigi Palmieri et al., replace these "hard" discrete constraints with convex inner approximations. This makes the safety requirements smooth and differentiable, allowing planners to steer trajectories away from obstacles before a collision even becomes possible. As Mario Gomez Andreu et al. show with their Field Overlap Collision Integral (FOCI), representing both the robot and the environment as continuous functions (like Gaussians) allows for high-speed navigation through tight spaces by treating collision avoidance as a continuous minimization problem.

Key Foundational Work

Separation-Sensitive Collision Detection for Convex Objects (1998) — contributed 1 concept elements to this direction.
Geometry-driven physical interaction between avatars and virtual environments (2004) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_collision_tunneling_40d542b7,
  title = {collision tunneling},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/collision-tunneling/rdr_40d542b74b2d42b59cc880563cc4b738},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_collision_tunneling_40d542b7_v1,
  title = {collision tunneling},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/collision-tunneling/rdr_40d542b74b2d42b59cc880563cc4b738/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake