Human-Centric Active Perception for Autonomous Observation

David Kent, Sonia Chernova

Abstract

As robot autonomy improves, robots are increasingly being considered in the role of autonomous observation systems -- free-flying cameras capable of actively tracking human activity within some predefined area of interest. In this work, we formulate the autonomous observation problem through multi-objective optimization, presenting a novel Semi-MDP formulation of the autonomous human observation problem that maximizes observation rewards while accounting for both human- and robot-centric costs. We demonstrate that the problem can be solved with both scalarization-based Multi-Objective MDP methods and Constrained MDP methods, and discuss the relative benefits of each approach. We validate our work on activity tracking using a NASA Astrobee robot operating within a simulated International Space Station environment.

Introduction

In high-stakes environments like the International Space Station (ISS) or deep-sea research stations, recording human activity is vital for documentation and later analysis. Traditionally, this requires a human operator or static cameras. As robots like NASA’s Astrobee (a free-flying cube) become more autonomous, they are being tapped to act as "autonomous cameramen."

Simulation of the NASA Astrobee robot observing a human within the International Space Station (ISS) environment.

However, filming a moving human in a 3D environment is complex. The robot must maximize "observation reward" (getting a clear, high-resolution shot of the task) while balancing three conflicting factors: safety (avoiding collisions), intrusion (minimizing noise and proximity to the human), and efficiency (managing limited battery life).

Methodology

The authors formalize the problem as a Semi-Markov Decision Process (SMDP). Unlike a standard MDP, an SMDP allows for actions that take variable amounts of time, such as moving between different modules of the ISS. Two primary optimization strategies are compared:

  • MOMDP with Scalarization: This method aggregates all rewards and costs into a single scalar value using a weighted sum. The robot then maximizes this value, solved via backwards induction.
  • Constrained MDP (CMDP): This method prioritizes a primary goal (maximizing filming quality) while treating other factors as "budgets" or hard limits (e.g., "do not exceed a specific intrusion threshold"). It is solved using Linear Programming.

To handle human unpredictability, human movement is modeled as a probability distribution based on previous task data rather than a static point.

Experimental Setup

The study utilized NASA’s Gazebo simulation of the ISS. Three distinct "human task" scenarios were created for evaluation: Experiment (fixed stations), Inspection (continuous slow movement), and Equipment Transfer (long-distance back-and-forth movement).

Visualizing the robot's planned trajectory (green arrows) relative to human activity zones.

Key Results and Discussion

Both methods successfully produced optimal policies that approached the "perfect knowledge" baseline. In all scenarios, the robots effectively tracked the human while respecting safety and power constraints.

Satisficing vs. Optimizing

A key insight is the human-centric benefit of CMDPs. In MOMDPs, developers must tune abstract weights, which is often unintuitive. CMDPs allow for satisficing—setting clear, interpretable budgets (e.g., "use no more than 40% battery"). This makes it significantly easier for human designers to reason about robot behavior. Additionally, CMDPs proved more robust when tasks changed; a budget remains a budget regardless of the specific task, whereas MOMDP weights often require re-tuning.

The Computational Trade-off

While CMDPs are easier to tune, they are computationally expensive. In tests, the MOMDP solved a 180-second task in 1 second, while the CMDP took a median of 237 seconds. This "computational wall" means CMDPs currently struggle with real-time reactive re-planning if a human behaves unexpectedly.

Conclusion and Future Work

The primary limitation remains the solve time for CMDPs. Future research will explore approximate CMDP solvers, such as Constrained Monte Carlo Tree Search, to combine the ease of tuning found in CMDPs with the speed of MOMDPs. Additionally, the researchers aim to transition from fully observable environments to those with higher uncertainty (POMDPs).

Summary updated

Abstract

As robot autonomy improves, robots are increasingly being considered in the role of autonomous observation systems -- free-flying cameras capable of actively tracking human activity within some predefined area of interest. In this work, we formulate the autonomous observation problem through multi-objective optimization, presenting a novel Semi-MDP formulation of the autonomous human observation problem that maximizes observation rewards while accounting for both human- and robot-centric costs. We demonstrate that the problem can be solved with both scalarization-based Multi-Objective MDP methods and Constrained MDP methods, and discuss the relative benefits of each approach. We validate our work on activity tracking using a NASA Astrobee robot operating within a simulated International Space Station environment.
BibTeX
@inproceedings{kent2020human,
  author = {David Kent and Sonia Chernova},
  title = {{Human-Centric Active Perception for Autonomous Observation}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2020},
  url = {https://openreview.net/forum?id=QYv_huzzwFV},
}
Figure