Wednesday 09 April 2025
The quest for socially aware robots that can navigate complex environments without bumping into humans or objects has been an ongoing challenge in robotics research. A new approach, however, shows promising results by incorporating human activities and preferences into the planning process.
The traditional method of motion planning involves creating a trajectory for a robot to follow, taking into account static obstacles and spatial constraints. However, this approach falls short when it comes to dynamic environments where humans are present. To address this issue, researchers have been exploring the use of 3D scene graphs (3DSGs), which provide a structured representation of the environment, including objects, spaces, and relationships.
In a recent paper, scientists from KTH Royal Institute of Technology and University of Stuttgart presented an innovative approach that integrates human activities and preferences into the planning process. The method starts by creating a 3DSG for a given scene, which includes nodes representing objects and edges defining spatial relationships between them. The researchers then manually introduce one or more humans into the scene, incorporating their activities and preferences into the graph.
The next step is to identify relevant objects along a planned trajectory that may impact the robot’s movement. This is done by searching for objects within a defined radius around each waypoint on the trajectory. For each object, the system generates a description including its ID, label, centroid, extents, affordances (the actions it allows), and attributes.
The enriched 3DSG is then fed into an LLM (Large Language Model), which assigns costs to each relevant object based on its impact factor in the trajectory. The cost reflects how much the object affects the robot’s movement, while a clearance value acts as a diminishing factor, reducing the impact as the robot moves farther from the object.
The computed costs are then integrated into a planner, generating an optimal trajectory that takes into account both human activities and preferences. This approach allows robots to adjust their movements in response to changing environments and human behavior, ensuring socially aware navigation.
The researchers evaluated their method using a scene with a human sitting on a bed watching TV. They compared the results with two baseline planners: one without human information and another that only considered spatial relationships. The preliminary findings show promising results, with the proposed approach producing more context-aware trajectories that respect human presence and preferences.
This innovative approach has significant implications for robotics research, enabling robots to navigate complex environments with greater ease and social awareness.
Cite this article: “Unlocking Human-Robot Harmony: A Novel Approach to Context-Aware Navigation”, The Science Archive, 2025.
Robotics, Motion Planning, 3D Scene Graphs, Human Activities, Preferences, Socially Aware Navigation, Dynamic Environments, Object Recognition, Trajectory Planning, Large Language Models
