Traditional robotic navigation often relies on a simplifying assumption: the robot is a point mass or a symmetric disk. While this abstraction simplifies the underlying math, it creates a "geometry gap" that prevents robots from navigating tight spaces. If a robot is modeled as a large bounding sphere, it will refuse to enter a narrow corridor even if it could easily fit by turning sideways.
Recent research has shifted toward geometry-aware motion planning, which explicitly incorporates a robot’s physical dimensions, contours, and orientation into the planning process. By treating the robot as a complex geometric volume rather than a dimensionless point, these systems can "thread the needle" through environments that were previously considered impassable.
The Limitation of Conservative Bounding Boxes
The primary reason for moving beyond point-mass models is efficiency. As Bernardini et al. et al. noted in early work on sweeper robots, failing to account for a robot's specific footprint leads to planning failures where the body might intersect obstacles during a turn. When robots are approximated by conservative bounding boxes or spheres, the navigable space is artificially restricted.
Gomez Andreu et al. et al. demonstrate that modeling a robot's specific geometry can effectively double the navigable space compared to traditional box-based methods. This is particularly critical in search and rescue (SAR) scenarios. Farinelli et al. et al. showed that incorporating physical dimensions allows autonomous systems to traverse passages only slightly wider than the robot itself, a necessity in cluttered disaster zones.
Orientation as a Navigational Tool
In geometry-aware planning, orientation ($SE(2)$ for ground robots or $SE(3)$ for aerial ones) is not just a result of movement—it is a tool for collision avoidance. By rotating its body, a robot can change its "effective footprint" relative to an obstacle.
For example, Long et al. et al. developed analytic methods to calculate the exact distance between polygonal robots and elliptical obstacles. By integrating orientation into Control Barrier Functions (CBFs), they enable robots to navigate narrow gaps by precisely timing their rotations. Similarly, Sarvaiya et al. et al. introduced PolyFly, which models quadrotors and their payloads as orientation-aware polytopes. This allows drones to tilt their way through vertical gaps that would be inaccessible to planners using orientation-invariant safety spheres.
This orientation awareness is also vital for landing. Gao et al. et al. argue that modeling a vehicle as a geometric volume in $SE(3)$ is essential for perching on inclined surfaces, as it prevents peripheral components like rotors from striking the platform during aggressive maneuvers.
Achieving Zero-Gap Precision
To make these complex models useful in real-time, researchers have developed differentiable representations of collision. Instead of checking for binary collisions, these methods calculate how much a shape must be scaled or moved to avoid contact.
- Scale Optimization: Wang et al. et al. proposed a "zero-gap" algorithm that transforms whole-body collision evaluation into a low-dimensional linear program. By calculating the minimum scale factor for two convex objects to collide, they achieve exact analytic gradients, allowing for aggressive maneuvers in highly constrained spaces.
- Differentiable Overlaps: Gomez Andreu et al. et al. use a differentiable "overlap integral" to represent both the robot and the environment as sets of Gaussians, enabling fast trajectory optimization directly within 3D environments.
- Swept Volumes: Li et al. et al. utilize swept-volume-based generation to refine initial paths. By considering the actual volume a robot occupies as it moves, their "coarse-to-fine" framework prevents the premature elimination of narrow but navigable routes.
Applications in Complex Environments
The shift toward geometry-aware planning has enabled new capabilities across various robotic platforms:
- Multi-Agent Systems: Wen et al. et al. and Andreychuk et al. et al. have extended these concepts to multi-agent pathfinding (MAPF). By accounting for rectangular body shapes and continuous-time volume, these planners allow for much tighter agent packing in warehouses than point-mass models.
- Autonomous Parking: Liu et al. et al. integrated vehicle contours and "expansion angles" into local navigation, allowing cars to execute precise turns in tight parking structures.
- Subterranean Flight: Wang et al. et al. demonstrated quadrotors navigating narrow, unstructured tunnels with clearances as small as 1.25 times the vehicle's diameter, outperforming human pilots by strictly adhering to geometric and aerodynamic constraints.
By moving away from conservative approximations and toward exact geometric representations, these researchers have enabled robots to operate with a level of spatial awareness that matches their physical reality.
Key Foundational Work
Autonomous exploration for search and rescue robots (2007) — contributed 1 concept elements to this direction.
Path-Planning for a Mobile Robot Sweeper (1994) — contributed 1 concept elements to this direction.