Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AVOCADO: Adaptive Optimal Collision Avoidance driven by Opinion

Diego Martinez-Baselga, Eduardo Sebastián, Eduardo Montijano, Luis Riazuelo, Carlos Sagüés, Luis Montano

TL;DR

AVOCADO addresses collision avoidance in crowds of mixed cooperative and non-cooperative agents by coupling a velocity-obstacle based planner with a nonlinear opinion-dynamics adaptive law that estimates each agent's degree of cooperation in real time from onboard perception. The method introduces per-agent variables o_i (through alpha_i = (o_i+1)/2), an attention A_i to modulate adaptation, and a projection-based estimator e_i to infer agent responses without communication. A linear program computes the closest collision-free velocity within per-agent admissible sets OCA_i, while attention-driven noise injection breaks symmetry deadlocks. Extensive simulations and real-world experiments with robots and humans demonstrate superior success rates, efficient paths, and robust zero-shot transfer, highlighting AVOCADO as a practical, low-cost solution for crowded environments. The approach is extensible to static obstacles and, with future work, to higher-level planning and 3-D settings.

Abstract

We present AVOCADO (AdaptiVe Optimal Collision Avoidance Driven by Opinion), a novel navigation approach to address holonomic robot collision avoidance when the robot does not know how cooperative the other agents in the environment are. AVOCADO departs from a Velocity Obstacle's (VO) formulation akin to the Optimal Reciprocal Collision Avoidance method. However, instead of assuming reciprocity, it poses an adaptive control problem to adapt to the cooperation level of other robots and agents in real time. This is achieved through a novel nonlinear opinion dynamics design that relies solely on sensor observations. As a by-product, we leverage tools from the opinion dynamics formulation to naturally avoid the deadlocks in geometrically symmetric scenarios that typically suffer VO-based planners. Extensive numerical simulations show that AVOCADO surpasses existing motion planners in mixed cooperative/non-cooperative navigation environments in terms of success rate, time to goal and computational time. In addition, we conduct multiple real experiments that verify that AVOCADO is able to avoid collisions in environments crowded with other robots and humans.

AVOCADO: Adaptive Optimal Collision Avoidance driven by Opinion

TL;DR

AVOCADO addresses collision avoidance in crowds of mixed cooperative and non-cooperative agents by coupling a velocity-obstacle based planner with a nonlinear opinion-dynamics adaptive law that estimates each agent's degree of cooperation in real time from onboard perception. The method introduces per-agent variables o_i (through alpha_i = (o_i+1)/2), an attention A_i to modulate adaptation, and a projection-based estimator e_i to infer agent responses without communication. A linear program computes the closest collision-free velocity within per-agent admissible sets OCA_i, while attention-driven noise injection breaks symmetry deadlocks. Extensive simulations and real-world experiments with robots and humans demonstrate superior success rates, efficient paths, and robust zero-shot transfer, highlighting AVOCADO as a practical, low-cost solution for crowded environments. The approach is extensible to static obstacles and, with future work, to higher-level planning and 3-D settings.

Abstract

We present AVOCADO (AdaptiVe Optimal Collision Avoidance Driven by Opinion), a novel navigation approach to address holonomic robot collision avoidance when the robot does not know how cooperative the other agents in the environment are. AVOCADO departs from a Velocity Obstacle's (VO) formulation akin to the Optimal Reciprocal Collision Avoidance method. However, instead of assuming reciprocity, it poses an adaptive control problem to adapt to the cooperation level of other robots and agents in real time. This is achieved through a novel nonlinear opinion dynamics design that relies solely on sensor observations. As a by-product, we leverage tools from the opinion dynamics formulation to naturally avoid the deadlocks in geometrically symmetric scenarios that typically suffer VO-based planners. Extensive numerical simulations show that AVOCADO surpasses existing motion planners in mixed cooperative/non-cooperative navigation environments in terms of success rate, time to goal and computational time. In addition, we conduct multiple real experiments that verify that AVOCADO is able to avoid collisions in environments crowded with other robots and humans.
Paper Structure (14 sections, 3 theorems, 21 equations, 18 figures, 1 table, 1 algorithm)

This paper contains 14 sections, 3 theorems, 21 equations, 18 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. Problem formulation
  4. AVOCADO: Adaptive Optimal Collision Avoidance driven by Opinion
  5. Optimal Collision Avoidance for Unknown Degrees of Cooperation
  6. Nonlinear Opinion Dynamics Adaptive Law
  7. Nonlinear Opinion Dynamics Adaptive Law Tuning
  8. Exploiting attention to avoid symmetry deadlocks
  9. Simulated results
  10. Head-on scenarios
  11. Multi-agent scenarios
  12. Static obstacles
  13. Experimental results
  14. Conclusions

Key Result

Proposition 1

Let $d_i>0$, $a_i>0$ and $b_i \in [-1, 1]$. Then, ${o}_i = b_i/d_i$ is an unstable equilibrium point of the nonlinear opinion dynamics adaptive law in eq:NODAL if $A_i>1/\left(a_i\text{sech}^2 \left(a_i \frac{b_i}{d_i}\right)\right)$.

Figures (18)

  • Figure 1: Illustrative example of one of the experiments with real robots and humans. Robots using AVOCADO adapt online to avoid collisions with the other entities in the arena despite not knowing the degree of cooperation of the other robots and humans. Section \ref{['sec:experiments']} discusses all the experiments in detail.
  • Figure 2: Geometry of $\mathsf{VO}_{i}$ (purple) and admissible velocities' regions associated to different degrees of cooperation (orange, green, blue). AVOCADO selects the closest velocity to the desired one that is inside all the admissible velocity sets.
  • Figure 3: Representation of the intersection set specified by the $\mathsf{OCA}_i$ half-planes of $4$ agents for different values of $\alpha_i$. Note that $\alpha_i$ may be different for each of the agents, but it is considered the same in this figure for a simple visualization. If the intersection set of all half-planes is not empty, the problem is guaranteed to have a solution.
  • Figure 4: Evolution of the nonlinear opinion dynamics adaptive law for a non-cooperative (top, orange) or a cooperative (bottom, blue) agent.
  • Figure 5: Geometry behind the projection estimator for ${e}_i$.
  • ...and 13 more figures

Theorems & Definitions (5)

  • Proposition 1
  • proof
  • Corollary 1
  • Proposition 2
  • proof