DOBB: Dynamic Oriented Bounding Boxes for Multi-Agent Collision Avoidance under State Uncertainty and Heterogeneous Dynamics
Organizations: Pedro Antonio Alarcon Granadeno is a PhD student in the Computer Science and Engineering department at the University of Notre Dame, IN, USA · Jane Cleland-Huang is a Professor in the Department of Computer Science and Engineering, University of Notre Dame, IN, USA
Abstract
Learning-based decentralized collision avoidance must remain reliable under state-estimation uncertainty and heterogeneous agent dynamics. Existing methods typically address these factors through training-domain variation or by exposing uncertainty and capability estimates to the policy at runtime. However, both leave the policy to infer how these quantities should alter the physical separation required for safe interaction, which can couple collision-avoidance behavior to relationships learned from the training distribution. We introduce the Dynamic Oriented Bounding Box (DOBB), an online geometric representation that maps the current state estimate, uncertainty, and agent-specific response characteristics into a dynamic, direction-dependent safety margin. We derive geometric features that jointly characterize the relative configuration and severity of interactions between agents' DOBBs, along with a reward function that preserves the physical interpretation of those interactions. Across four evaluation environments, DOBB achieves the lowest mean total failure rate of 2.6%. This performance persists under out-of-distribution shifts in actuation dynamics and state-estimation uncertainty, with collision-based failure remaining at or below 2.3% in every tested condition. We attribute this robustness to the geometric abstraction of the safety margin, which is less coupled to the specific uncertainty and actuation regimes encountered during training.
Figures & tables
| Condition | Geometric interpretation |
|---|---|
| lies entirely beyond host face , so and are disjoint ( ). gives normalized projected clearance from face ; multiple positive components indicate compound placement, e.g., front-left. | |
| A boundary of the projection coincides with the supporting line of host face . If , another SAT axis separates and ; if , they are in boundary contact involving face . In either case, . | |
| The projections of and overlap along both host axes. indicates separation along a neighbor-aligned SAT axis, boundary contact, and strict intersection. For , gives the minimum separating translation and quantifies normalized encroachment. |
| Parameter | Training range |
|---|---|
| Preferred speed | |
| Physical radius | |
| Maximum acceleration | |
| Maximum jerk | |
| Maximum yaw rate | |
| Maximum yaw acceleration |
| Environment | Constrained dynamics | Heterogeneous capabilities | Noisy sensing and EKF |
|---|---|---|---|
| – | – | – | |
| – | – | ||
| – | |||
| Policy | Training environment | Observations | Reward |
|---|---|---|---|
| [-1pt] | Eqs. 14 – 15 | Eq. 17 | |
| Eqs. 14 – 15 & Eqs. 18 – 19 | Eq. 17 | ||
| Eq. 16 | Eq. 17 | ||
| Eq. 16 | Eq. 13 |
| Policy | [-1pt] | ||||
|---|---|---|---|---|---|
| [ 3 ] | [-1pt] 1.0 /0.8/0.0 | [-1pt] 69.3/3.1/8.5 | [-1pt] 69.7/3.8/9.6 | [-1pt] 71.0/3.8/10.4 | [-1pt] 52.8 |
| [-1pt] 2.5/20.6/0.8 | [-1pt] 6.3/5.0/0.7 | [-1pt] 7.3/4.6/1.3 | [-1pt] 9.3/2.1/0.9 | [-1pt] 6.4 | |
| [-1pt] 1.1/51.8/1.2 | [-1pt] 5.3/5.6/0.4 | [-1pt] 6.5/5.5/1.2 | [-1pt] 8.1/4.3/0.7 | [-1pt] 5.3 | |
| [-1pt] 3.6/5.2/0.7 | [-1pt] 5.3/4.2/0.8 | [-1pt] 7.4/3.3/0.7 | [-1pt] 7.9/2.0/0.6 | [-1pt] 6.1 | |
| [-1pt] 10.5/10.3/1.7 | [-1pt] 2.2/1.5/0.2 | [-1pt] 4.7/1.2/0.3 | [-1pt] 4.5/0.7/0.1 | [-1pt] 5.5 | |
| [-1pt] 12.8/0.2/0.8 | [-1pt] 3.3/0.3/0.1 | [-1pt] 3.9/0.3/0.3 | [-1pt] 3.2/0.2/0.3 | [-1pt] 5.8 |
Collision Avoidance
Collision avoidance develops methods for predicting and preventing unsafe encounters among autonomous agents, vehicles, robots, and obstacles.
See all 98 papers