CoBots: Robust Symbiotic Autonomous Mobile Service Robots

Manuela M. Veloso, Joydeep Biswas, Brian Coltin, Stephanie Rosenthal

Introduction

The CoBot (Collaborative Robot) project at Carnegie Mellon University represents a significant milestone in long-term autonomous service robotics. Over several years of deployment, a fleet of CoBots has successfully navigated more than 1,000 km within multi-floor office buildings. Unlike robots designed for isolated laboratory settings, CoBots are engineered to handle the inherent messiness and unpredictability of real-world human environments. This robustness is achieved through a combination of innovative localization techniques, intelligent multi-robot scheduling, and a unique "symbiotic" relationship with human occupants.

The CoBot fleet at Carnegie Mellon University.

Challenges in Human Environments

Autonomous service robots typically face two primary hurdles: environmental dynamism and physical constraints.

  1. Dynamic Environments: Traditional localization algorithms often rely on static maps. In active offices, however, furniture is moved, boxes are stacked, and people are constantly in motion. These "short-term features" (STFs) introduce noise that can cause standard Markov-based localization to fail.
  2. Physical Limitations: Performing complex manipulation tasks, such as pressing elevator buttons or handling mail, requires expensive and heavy robotic arms. Building a fully independent robot capable of these tasks is often cost-prohibitive and technically daunting.

The CoBot project addresses these challenges by moving away from the ideal of total independence, focusing instead on a "symbiotic autonomy" model where the robot is robust enough to navigate autonomously but intelligent enough to request human assistance for tasks beyond its physical capabilities.

Technical Pillars

Episodic non-Markov Localization (enML)

To handle dynamic environments, the researchers developed Episodic non-Markov Localization (enML). Traditional Markov localization assumes that the current state depends only on the immediate previous state. In contrast, enML utilizes a "Varying Graphical Network" (VGN) to track three distinct feature types: Long-Term (permanent structures like walls), Short-Term (movable furniture), and Dynamic (moving people).

The enML framework treats navigation as a series of episodes, allowing the robot to reason about short-term features without computational bloat.

By treating navigation as a sequence of "episodes," the robot can maintain internal consistency regarding unmapped objects while discarding stale data once it reaches a point of independence. This approach allows the robot to treat moved furniture as temporary landmarks rather than sensor noise, significantly improving reliability in changing spaces.

Multi-Robot Scheduling with Transfers

Managing a fleet requires sophisticated coordination. The CoBot scheduler supports time and location constraints and introduces the concept of transfers. For example, one robot can retrieve a package on an upper floor and hand it to another robot at the elevator. This allows robots to specialize in certain areas or floors, optimizing the overall efficiency of the fleet. The system employs Mixed Integer Programming (MIP) for small task sets and Traveling Salesperson Problem (TSP) approximations for larger scales.

Symbiotic Autonomy

Symbiotic autonomy is a planning framework where "asking for help" is treated as a primary action. When a CoBot needs to change floors, it navigates to the elevator and proactively asks a nearby human to press the button. The planner uses a decision-theoretic model to select the best person to ask, considering factors like distance, the person's current activity, and how frequently they have been interrupted previously.

Evaluation and Results

The system was evaluated through the "1,000km Challenge" at CMU’s Gates-Hillman Center. Key findings include:

  • Localization Robustness: The use of enML increased the mean distance between human interventions from 4.79 km to 8.13 km. In challenging open areas like atria, the improvement was even more significant, jumping from 0.62 km to 4.42 km.
  • Operational Efficiency: Proactively seeking human help for elevator operation reduced task completion times by an average of 4 minutes compared to passive waiting.
  • User Integration: During a two-week trial, over 100 occupants successfully used the robots for tasks such as snack delivery, mail transport, and visitor escorting.

Conclusion

The CoBot project demonstrates that perfect independence is not a prerequisite for effective service robotics. By combining robust localization that accounts for environmental change with a symbiotic approach to human-robot interaction, the CoBots have proven their ability to provide reliable service in complex, everyday environments. Future research will focus on learning human preferences to minimize social friction and improving natural language interfaces for more intuitive interaction.

Summary updated

Abstract

We research and develop autonomous mobile service robots as Collaborative Robots, i.e., CoBots. For the last three years, our four CoBots have autonomously navigated in our multi-floor office buildings for more than 1,000km, as the result of the integration of multiple perceptual, cognitive, and actuations representations and algorithms. In this paper, we identify a few core aspects of our CoBots underlying their robust functionality. The reliable mobility in the varying indoor environments comes from a novel episodic non-Markov localization. Service tasks requested by users are the input to a scheduler that can consider different types of constraints, including transfers among multiple robots. With symbiotic autonomy, the CoBots proactively seek external sources of help to fill-in for their inevitable occasional limitations. We present sampled results from a deployment and conclude with a brief review of other features of our service robots.
BibTeX
@inproceedings{veloso2015cobots,
  author = {Manuela M. Veloso and Joydeep Biswas and Brian Coltin and Stephanie Rosenthal},
  title = {{CoBots: Robust Symbiotic Autonomous Mobile Service Robots}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2015},
}
Figure