Research Work

The Art of Navigation: Strategies in Crowded Spaces

Introduction:

 Navigating in crowded spaces presents a formidable challenge, not only for the burgeoning field of robotics but also within the natural world. In robotics, navigation involves planning algorithms and control systems that manage the robot’s motion from one point to another. Traditionally, this process is decoupled into two phases: motion planning and subsequent execution. This method, although effective in structured and predictable environments, often falters in the face of the unpredictable and dynamic nature of crowded spaces.  Conversely, natural organisms exhibit exceptional abilities to navigate these complex environments with innate fluidity and adaptiveness. The agility of a fox darting through a dense forest, the synchronized swarming of fish in coral reefs, or the bustling movement of humans across busy urban intersections—all demonstrate sophisticated biological systems of navigation that integrate sensory data, cognitive processing, and physical movement in real-time.

Understanding Decoupled Motion Planning and Control

Motion planning involves generating a reference path or trajectory from an initial state to a goal state, considering various constraints (temporal and spatial). This phase is typically followed by the control design phase, where feedback control mechanisms are employed to ensure that the robot adheres to the planned paths/trajectories. It has clear structural advantages, such as simplifying complex problems into manageable sub-parts. However, it introduces a distinct separation between planning and control execution.

Natural Locomotion

 In nature, organisms exhibit an intricate blend of sensing, processing, and acting that allows for seamless movement within their environments. For example, a cheetah chasing prey adjusts its motion in real-time, integrating immediate sensory inputs with motor execution. Other examples include birds in flight continuously adjusting their natural trajectory in response to wind currents and obstacles using a complex interplay of neural feedback and muscular control. This allows for complex maneuvers that are highly adapted to the immediately observed environment. The seamless integration of sensory inputs and motor responses enables biological organisms to exhibit robust locomotion strategies that are highly responsive and adaptable to sudden environmental changes.

Limitations (Decoupled Approaches):

 Decoupled approaches in robotics struggle with dynamic environments where conditions can change unpredictably. The pre-computed paths/trajectories may not account for new obstacles or alterations in the terrain, leading to potential delays as the system attempts to recalculate its reference trajectories (sometimes even fails to adapt swiftly enough). The main limitations stem from the inherent latency between the planning phase and execution. Any delay in feedback integration can lead to discrepancies between the robot's motion and the actual requirements for navigation or task performance. Moreover, the rigidity of precomputed plans does not allow for the level of flexibility often required in crowded, real-world scenarios. Therefore, despite their technical merits such as simplifying complex tasks into manageable segments, decoupled strategies have a clear divide between planning and execution, potentially leading to subpar performance in uncontrolled dynamic environments such as crowds.

Example of limitation in robotics: A robot navigating a precomputed path in a warehouse may fail to adapt quickly to an unexpected obstacle like a fallen object, requiring a stop and re-computation of the path. In contrast, a squirrel darting across branches does not precompute its path but dynamically adjusts its trajectory based on immediate sensory feedback, showcasing a remarkable ability to handle unpredictability. Natural systems are inherently designed to handle such unpredictabilities through continuous feedback control mechanisms.

Key Philosophy:

 The key philosophy of my research work revolves around developing highly responsive novel navigation frameworks for robotic systems that couple the notions of planning, control, and safety (based on local sensing) as observed in nature to enable robotic systems to be deployed in dynamic/unpredictable (difficult to predict) spaces. A classic example of such spaces are crowds where motion prediction of environmental entities during run-time is difficult and can often only be determined with uncertainties.

 Safe navigation in crowded spaces presents a unique set of challenges. Crowded spaces are complex, dynamic and require sensing the dynamic bounds of safety during run-time. Motion prediction of dynamic environmental entities during operation is extremely hard, and the unstructured, unknown nature of the operating spaces presents challenges in localization and mapping. Guarantees of safety are vital when human subjects share the operating space, and rapid control computations are essential to deal with ever-changing scenarios.

 In this context, motion sets (from non-linear feedback control designs) are identified as fundamental building blocks for integrated navigation frameworks (in the figure below). They provide estimated bounds on naturally induced motion when invoking a feedback controller. Such sets have well-defined geometric structures and are a natural consequence of the non-linear feedback control designs. They can be thought of as the physical manifestation of the feedback controller itself. These motion sets are directly employed to induce naturally safe system trajectories.

The key objectives in the proposed navigation framework are as follows: