Joint Communication and Motion Planning for Cobots

Mehdi Dadvar, Keyvan Majd, Elena Oikonomou, Georgios Fainekos, Siddharth Srivastava

TL;DR

This paper introduces a unified framework that allows collaborative robots (cobots) to simultaneously plan their physical movements and their explicit communication signals (like "I'm turning left"). By modeling how humans perceive these signals and adjust their own paths, the system avoids the "robot freezing problem" and resolves deadlocks in tight spaces where movement alone isn't enough to signal intent.

What Problem Does This Paper Address?

Navigating a shared workspace with humans is a major challenge for autonomous robots. Traditional motion planners often treat humans as dynamic obstacles to be avoided. This leads to two main issues:

  1. The Robot Freezing Problem: In confined spaces (like a narrow hallway), a robot might find no "safe" path because any movement appears to risk a collision, causing the robot to stop indefinitely.
  2. Ambiguity: Relying solely on "implicit communication" (the robot's physical trajectory) can be misleading. A human might not understand a robot's intent until it's too late to avoid a conflict.

The authors argue that for efficient co-working, robots need to use explicit communication (digital signals, lights, or voice) to influence human behavior and create mutually feasible paths.

Methods

The researchers propose a Deliberative Communication Planning (PDC) framework consisting of two main components:

  • Motion Planner (MP): This uses a modified version of CBF-TB-RRT (Control Barrier Function enabled Time-Based Rapidly-exploring Random Tree). It generates a library of diverse, collision-free path segments that satisfy safety constraints defined by Control Barrier Functions—a mathematical tool that guarantees the robot stays within a "safe set" of states.
  • Communication Planner (CP): This module builds a search tree to evaluate combinations of motion plans and communication signals (e.g., "Going North," "Staying Silent").
  • Belief Modeling: The CP maintains a "belief" model of what the human thinks the robot will do. It uses a noisy sensor model to account for the fact that humans might misinterpret or miss a robot's signal.
  • Joint Optimization: It selects the pair of (action, signal) that minimizes a multi-objective cost function. This function balances the robot's travel time, the human's predicted travel time, the proximity between them (safety), and the "cost" of communicating.

Datasets

The paper does not use external datasets. Instead, the authors evaluated their framework using custom-designed simulation environments built to test specific interaction challenges:

  • Basic: A spacious floor map for general navigation.
  • Hallway: A restricted environment to test head-on encounters.
  • Intersection: A confined space where only one agent can pass at a time, specifically designed to trigger deadlocks.

Human behavior in these simulations was modeled using the Social Forces Model (SFM) and the Dynamic Window Approach (DWA).

Key Insights and Ideas

  • Communication as a Path-Opener: Explicit communication isn't just a courtesy; it's a planning tool. By signaling intent, the robot "updates" the human's belief, prompting the human to move in a way that opens up a previously "blocked" path for the robot.
  • Handling Imperfect Perception: The model assumes humans aren't perfect sensors. By incorporating a noisy sensor model, the robot plans more robustly, choosing signals that are likely to be understood even in distracting environments.
  • Flexible Prioritization: The framework allows designers to tune weights in the cost function. You can prioritize the robot's efficiency (industrial setting) or the human's comfort and speed (service/hospitality setting) without changing the underlying algorithm.
  • Reducing Computational Overhead: By selecting a small set of "diverse" plans from the RRT rather than evaluating every possible path, the system remains computationally efficient enough for real-time use.

Key Results

The authors compared their joint approach against a baseline (CBF-TB-RRT without explicit communication):

  • Deadlock Resolution: In the "Intersection" and "Hallway" maps, the baseline frequently resulted in deadlocks or infinite proximity costs. The proposed method resolved these 100% of the time.
  • Efficiency: In the "Basic" environment, the joint planner reduced the number of planning iterations from an average of 46–113 down to just 2, indicating a much more stable and "decisive" behavior.
  • Safety: The Proximity Cost (PC)—a measure of how uncomfortably close the agents get—was dramatically lower in the joint framework across all test cases.
  • Speedup: When the robot was prioritized ($F=1$), its normalized speed increased by up to 2.7x compared to scenarios where the human was prioritized, demonstrating the framework's flexibility.

Limitations and Future Work

  • Model Dependency: The framework relies on having a reasonably accurate model of how humans move and react to signals. In the real world, human behavior is highly stochastic and varied.
  • Scalability: The current study focuses on one-on-one (one robot, one human) interactions. Scaling this to a crowded warehouse with multiple robots and dozens of humans would significantly increase the complexity of the belief-space reasoning.
  • Signal Modality: The paper treats communication signals as abstract symbols. Future work could explore specific modalities like Augmented Reality (AR), projection mapping, or natural language to see how they map to the "noisy sensor" model used here.

Summary updated

Abstract

The increasing deployment of robots in co-working scenarios with humans has revealed complex safety and efficiency challenges in the computation of the robot behavior. Movement among humans is one of the most fundamental —and yet critical—problems in this frontier. While several approaches have addressed this problem from a purely navigational point of view, the absence of a unified paradigm for communicating with humans limits their ability to prevent deadlocks and compute feasible solutions. This paper presents a joint communication and motion planning framework that selects from an arbitrary input set of robot's communication signals while computing robot motion plans. It models a human co-worker's imperfect perception of these communications using a noisy sensor model and facilitates the specification of a variety of social/workplace compliance priorities with a flexible cost function. Theoretical results and simulator-based empirical evaluations show that our approach efficiently computes motion plans and communication strategies that reduce conflicts between agents and resolve potential deadlocks.
BibTeX
@inproceedings{dadvar2022joint,
  author = {Mehdi Dadvar and Keyvan Majd and Elena Oikonomou and Georgios Fainekos and Siddharth Srivastava},
  title = {{Joint Communication and Motion Planning for Cobots}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2022},
  url = {https://openreview.net/forum?id=5cGyjCRTQ1a},
}

Loading related content...

Figure