Safe Navigation of Mobile Robots

Navigating indoor social environments presents challenges to mobile robots, frequently navigating dynamic obstacles amid static structures like corridors. Balancing efficient and safe path and motion planning remains a challenge due to their inherent a trade-off relationship. For instance, prioritizing safety may involve slower navigation, potentially compromising efficiency, such as travel time. My research aimes at maintaining safety while optimizing efficiency in these intricate environments.

Path Planning With Hierarchical Topology Map (HTM):

In mobile robot navigation, two primary tasks include global path planning and local motion planning. Global path planning typically considers static obstacles, as dynamic ones might not be visible to the robot beyond corridors, leaving handling of moving obstacles to the local planning stage.

When creating a global path, wider hallways should be preferred to enhance navigation speed with fewer collision concerns compared to narrow corridors. I incorporated this factor into global path planning by adjusting the maximum navigation speed based on corridor widths. This speed limit is determined by the robot's capacity to decelerate and stop. Wider corridors result in faster travel and shorter travel times, influencing the lower cost of the path. The image below depicts narrow corridors restrict speed increase as proximity to walls increases. Minimizing this cost yields the optimal path.

(pic) allowable navigation speed

Preferring wider hallways may offer qualitative benefits when encountering moving obstacles. While the robot cannot guarantee the absence of moving objects beyond its sight, wider corridors are less likely to be crowded with dynamic objects during the navigation. Consequently, the robot is more likely to achieve the speed anticipated at the moment of global path planning.

Considering navigation from a human perspective, paths emerge in a topological manner (e.g., home to bus stop: home-street A-street B-bus stop). Similarly, a suggested topology map and planning method use summed navigation costs of corridors mapped with the topology based on the skeleton of occupancy maps (an abstract map representation). The skeleton of a map and its topology form the Hierarchical Topology Map (HTM). 

(pic) HTM definition

Paths are planned on the topology of HTM, identifying corridors with minimum cost. 

(pic) skeleton path

 

Once corridors are planned, the HTM structure allows for the immediate retrieval of the skeleton path along with the corridors.

(Pic) Optimization of the corridors

Then, the planned skeleton path undergoes dynamic programming within discretized spaces along the selected corridors to eliminate unnecessary detours of the skeleton path. Real robot experiments verified these refined paths, highlighting the effectiveness of preferring wider corridors to increase the robot's navigation speed safely.

(GIF) Experiment HTM-EC

(see the video in Youtube)

(pic) experimental speed comparison

Moreover, as the proposed method relies on the map skeleton, it ensures completeness in path planning, meaning the guarantee in finding a feasible path, whenever it exists, unlike popular probabilistic methods such as RRT and PRM. Path planning exclusively using HTM, without conversion to an occupancy grid, marks another advantage. Comparisons in computation time demonstrate the method's computational efficiency, showing less variations across various map sizes and complexities.

(Pic) Path Planning with Various Maps

(Pic) Computation Comparison

Collision Avoiding Motion Planning With Collision Arc:

A mobile robot often encounters decision-making agents, where their interactions are crucial in determining their motion. Ignoring these interactions in motion planning can lead to the freezing robot problem (FRP), degrading planning efficiency. However, considering these interactions in far-sighted planning entails a high computational burden for predicting evolving agent trajectories. Although deep learning approaches mitigate computational demands, ensuring safety remains a challenge.

Given that humans are prevalent among dynamic objects in robot's surroundings and are adept at far-sighted and efficient navigation, integrating their navigation behavior model into the robot navigation appears promising. Thus, we aimed at developing a human-model based local planning method that prioritizes safety as an additional measure.

 

(Pic) Human Model (2011 Moussaid)

The pedestrian behavior model assumes humans prioritize direction changes over speed alterations unless essential. Visual information plays a primary role in their navigation, and the LiDAR data of the robot resembles such the human visual perception. Humans tend to navigate towards directions that minimize the distance to their goal, and they consider others' velocities into current visual measurements, forming the cognitive distance to the collision.

(pic) cognitive distance 2 agent

Under constant navigation speed as suggested by the model, the robot's velocities are mapped into a circle in velocity space, named as the Preferred Velocity (PV) circle. The Velocity Obstacle (VO) concept becomes involves to define the Collision Arc (CA) as the intersection with the VO and PV. Safety guarantee lies in selecting the angles outside the Collision Arc, supported by studies on the Velocity Obstacle proving collision-free velocity selections outside the VO. This finding can be added to the human model, assuring the safety while the efficient navigation characteristics hold. 

(pic) collision arc

This concept extends to encompass general object shapes and multiple objects in the environment, and was verified with the simulations. 

(pic) collision arc extension to general shape and multi body

(GIF) CA-4 agent simulation (pic) CA-4 agent with VO visulization(GIF) CA-10 agent simulation(GIF) CA-FRP