Asymmetric Responsibility Allocation for Multi-Agent Collision Avoidance

12 concepts8 papers1 proposal

When multiple autonomous agents—such as drones, self-driving cars, or mobile robots—navigate a shared space, they must coordinate to avoid collisions. The traditional paradigm relies on symmetric reciprocity: every agent splits the responsibility for safety equally. For example, if two agents are on a collision course, standard reciprocal velocity obstacle frameworks expect each agent to alter its velocity by exactly half of the required correction.

While elegant in theory, symmetric responsibility breaks down in the real world due to heterogeneity. Agents differ in their physical capabilities, situational risks, and social intents. Forcing a heavy, slow-turning automated truck and an agile quadcopter to split safety maneuvers 50/50 is highly inefficient, and at worst, physically impossible.

To solve this, a powerful research direction has emerged: asymmetric responsibility allocation. By treating collision avoidance as a dynamic, shiftable burden rather than a rigid, mutual constraint, multi-agent systems become significantly safer, more agile, and free from coordination deadlocks.

Factoring in Physical Agility and Kinematics

The most intuitive reason to break safety symmetry is that different robots have different physical limits. If a highly agile robot encounters a sluggish one, the agile robot should logically bear more of the evasion burden.

Wang et al. formalized this intuition by integrating heterogeneous dynamics into Control Barrier Functions (CBFs). In their framework, safety barrier certificates—which define the admissible control space to guarantee collision-free operation—are explicitly scaled by each robot's acceleration limits. An agent with higher acceleration capabilities is assigned a larger share of the safety constraint, forcing it to deviate more proactively, while the less agile agent can maintain its preferred trajectory undisturbed. This decentralized approach ensures safety even when robots have no prior knowledge of their neighbors' specific capabilities, minimally disrupting their primary tasks (Safety Barrier Certificates for Collisions-Free Multirobot Systems).

Symmetric Avoidance (Fails if Agent A is slow):
[Agent A (Heavy)]  --->  |  <---  [Agent B (Agile)]
                         ^ Collision Point (Equal deviation expected)

Asymmetric Avoidance (Succeeds):
[Agent A (Heavy)]  --------======> (Maintains path)
                             \
                              \___ [Agent B (Agile)] (Takes 100% responsibility)

Physical asymmetry is also vital when robots interact with humans. For instance, Luo et al. developed the PORCA model to govern autonomous vehicles navigating dense pedestrian crowds. A pedestrian is holonomic (can step sideways instantly), whereas a car is non-holonomic (constrained by a turning radius and momentum). By dynamically shifting a higher share of the collision avoidance responsibility to the pedestrian model during close-proximity encounters, the system avoids making unrealistic assumptions about the vehicle's stopping capabilities, resulting in smoother and more natural navigation. This stands in stark contrast to simpler pathfinding models that assume uniform, synchronized capabilities across all agents (Yakovlev et al.).

Balancing Situational Risk and Maneuvering Freedom

Responsibility should not just depend on static hardware limits; it must also adapt to an agent's immediate environment. If an otherwise agile robot is backed into a corner or navigating a narrow corridor, it loses its maneuvering freedom.

To address this situational imbalance, Lyu et al. and Dolan et al. proposed a decentralized safe control framework driven by dynamic risk allocation. Instead of a fixed safety split, responsibility shares are calculated online and are inversely proportional to the local risk each agent faces.

When an agent experiences high environmental pressure (e.g., navigating close to obstacles), its safety constraints are mathematically relaxed, granting it more autonomy and decision-making freedom to escape the high-risk zone. Conversely, the neighboring agent operating in a safer, open area receives a larger responsibility share. This safer agent's control space is tightly constrained, forcing it to actively carve out a safe path for both.

Resolving Deadlocks via Social Preferences

Beyond physics and risk, asymmetric responsibility serves as a powerful tool for breaking behavioral symmetry. When two identical agents encounter each other head-on with symmetric rules, they often suffer from "reciprocal dance" oscillations or deadlocks, where both repeatedly choose the same evasion side, mirror each other, and freeze.

Dolan et al. resolved this by introducing Responsibility-associated Social Value Orientation (R-SVO). By mapping an agent's underlying "personality"—ranging from egoistic to altruistic—to its safety responsibility, agents naturally break symmetry. An altruistic agent willingly absorbs a higher share of the safety burden, while an egoistic agent holds its ground. This diversity in roles prevents oscillatory behaviors, leading to a 46.8% improvement in deadlock resolution and significantly faster task completion times.

In game-theoretic motion planning, this concept can be framed as a continuous variable rather than a binary choice. Laine et al. demonstrated that by treating collision avoidance as a shiftable responsibility weight within a trajectory optimization objective function (a "politeness" term), an autonomous vehicle can smoothly interpolate between assertive and cautious behaviors.

By tracking multiple hypotheses of a neighboring driver's intent, the vehicle can dynamically adjust its responsibility allocation:

  1. High Confidence in Deference: If the vehicle senses the other driver is yielding, it reduces its own responsibility weight, executing an assertive maneuver.
  2. High Confidence in Aggression: If the vehicle senses the other driver will not yield, it increases its own responsibility weight, smoothly transitioning to a proactive, defensive trajectory.

By moving away from rigid, shared safety requirements and toward a fluid, asymmetric delegation of responsibility, modern multi-agent systems achieve a level of coordination that mirrors the adaptive, intuitive nature of human interaction.

Key Foundational Work

Safety barrier certificates for heterogeneous multi-robot systems (2016) — contributed 3 concept elements to this direction.
Safety Barrier Certificates for Collisions-Free Multirobot Systems (2017) — contributed 2 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_asymmetric_responsibility_allocation_677c31e3,
  title = {asymmetric responsibility allocation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/asymmetric-responsibility-allocation/rdr_677c31e314aa46c485e993f1318352af},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-22},
}

This exact version (version 1, frozen text):

@misc{lacuna_asymmetric_responsibility_allocation_677c31e3_v1,
  title = {asymmetric responsibility allocation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/asymmetric-responsibility-allocation/rdr_677c31e314aa46c485e993f1318352af/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-22},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake