TL;DR
This paper introduces a novel architecture for mobile manipulators designed for human-robot collaboration. It presents two key components: a "Human Aware Manipulation Planner" (MHP) that generates safe and comfortable motion paths considering human presence, and a "SHARY" supervision system that manages collaborative tasks by monitoring human commitment and communicating intentions.
What Problem Does This Paper Address?
The paper tackles the significant challenge of enabling robots to effectively and safely collaborate with humans on shared tasks. Traditional robots often operate in isolation due to safety concerns, limiting their potential for interaction. However, for true collaboration, robots need to:
- Understand and predict human behavior: This includes not just their physical presence but also their intentions, preferences, and comfort levels.
- Generate legible and predictable actions: Humans need to understand what the robot is doing and why to build trust and facilitate smooth interaction.
- Ensure safety and comfort: The robot's movements must not only avoid physical collisions but also respect human personal space and avoid causing discomfort.
- Manage task execution dynamically: Collaborative tasks are inherently dynamic, requiring robots to adapt to human actions and communicate their own status and intentions.
The authors aim to bridge this gap by developing a system that integrates human awareness into both the robot's motion planning and its overall task supervision.
Methods
The paper proposes a three-level architecture (hardware, functional, decisional) for human-robot interaction, exemplified by their robot platform, Jido. The core contributions lie in two modules:
- Human Aware Manipulation Planner (MHP): This module goes beyond simple collision avoidance. It generates motion paths for the robot's arm and base that consider:
- Safety: Ensuring a safe distance from the human.
- Visibility: Positioning the object for easy human viewing.
-
Arm Comfort: Minimizing the human's effort to reach for the object. It achieves this by defining cost functions for these properties and using an A* search on a 3D grid to find optimal object handling positions and paths. Generalized Inverse Kinematics is used to adapt the robot's arm motion to the object's planned path.
-
SHARY Supervision System: This system manages collaborative task execution. It operates on a "task and robot knowledge data base" and a "task refinement and execution engine." Key aspects include:
- Communication Acts: Defining a set of communicative acts (e.g.,
ask-task,propose-plan,realize-task,give-up) that facilitate explicit communication between human and robot. - Communication Schemes: Establishing predefined sequences of communication acts to manage task flow and expectations.
- Monitoring: Actively monitoring human commitment (e.g., by tracking face direction and arm movement) to ensure synchronized behavior and detect deviations.
- Task Decomposition: Breaking down high-level tasks into atomic sub-tasks, including those for monitoring human behavior.
Datasets
Not discussed in the paper. The paper focuses on the architecture and algorithms, with demonstrations using their Jido robot platform.
Key Insights and Ideas
- Human-centric motion planning is crucial: Simply avoiding collisions isn't enough for collaboration. Considering human comfort, visibility, and preferences leads to more natural and acceptable robot behavior.
- Explicit communication is key for collaboration: Defining a set of communication acts and schemes allows robots and humans to clearly signal intentions, plans, and task status, fostering mutual understanding and trust.
- Monitoring human commitment is essential: Robots need to actively observe human actions and infer their engagement in a collaborative task to adapt their own behavior and ensure synchronization.
- Integrating motion planning with task supervision creates a more robust system: By linking the MHP's motion generation with SHARY's task management, the robot can react intelligently to human actions and task progress.
Key Results
The paper demonstrates the effectiveness of their approach through a "Bring-Object" collaborative task scenario.
- Successful Grasping Case: The robot successfully navigates, picks up an object, and presents it to the human in a safe and comfortable manner. The human's action of grasping is interpreted as an
End-Taskcommunication act. - Unsuccessful Grasping Case: When the object is not placed in the gripper, simulating a pick-up error, SHARY correctly interprets this as a task failure. It then executes a
Give-Upcommunication act, informing the human of the failure and closing the interaction loop.
While quantitative results like specific accuracy metrics are not detailed, the qualitative demonstration highlights the system's ability to handle both successful and erroneous scenarios in a collaborative context.
Limitations and Future Work
- Limitations:
- The current implementation relies on specific communication acts and schemes, which may need to be expanded for more complex interactions.
- Speech recognition for human input was not fully integrated at the time of writing, relying on implicit "YES" responses.
- Autonomous grasping was not fully robust, requiring manual placement of the object in the gripper in some demonstrations.
- Future Work:
- Integrating on-line planning abilities that explicitly consider Human-Robot Interaction and Collaborative Problem Solving.
- Using probabilistic methods to ensure continuity between base and arm motion for more fluid combined movements.
- Extending the system to handle scenarios with multiple humans and multiple concurrent tasks.
- Strengthening the link between motion planners and supervisors for even more sophisticated robot reactions to human states.
Summary updated