SHIFT Planner

Achieving efficient and uniform coverage in obstacle-laden unknown environments is essential for autonomous robots in cleaning, inspection and agricultural operations. Unlike most existing approaches that prioritize path length and time optimality, we propose the SHIFT Planner framework, which integrates semantic mapping, adaptive coverage planning, and real-time obstacle avoidance to ensure comprehensive coverage across diverse terrains and semantic features. Key innovations include:

• Radiant Field-Informed Coverage Planning (RFICP) algorithm: Generates trajectories that adapt to terrain variations by aligning with environmental changes. Additionally, a Gaussian diffusion field is employed to ensure efficient and uniform coverage under varying environmental conditions influenced by target semantic attributes.
• Incremental IKD-tree Sliding Window Optimization (IKD-SWOpt): Optimizes trajectory segments within and outside waypoint safety zones by evaluating and refining non-compliant segments through an adaptive sliding window approach. This method not only reduces computational overhead but also guarantees the quality of real-time obstacle avoidance trajectories.

Overall architecture

Overall architecture of the SHIFT planner. The terrain-derived point cloud data is initially filtered and integrated into a global map. This map, combined with a semantic map, serves as input for the global planner. RFICP then generates an adaptive coverage trajectory by adjusting the robot’s speed based on semantic information, while IKD-SWOpt performs real-time refinements of local trajectories when obstacles are encountered

RFICP: Gaussian Diffusion Kernel

RFICP: Speed Adaptive Coverage Path

RFICP: Boustrophedon Coverage Path

Sequence the adjusted landmark points in a boustrophedon (zigzag) pattern to ensure systematic and complete coverage of the terrain.

RFICP Pipeline

RFICP generates a terrain-conforming coverage trajectory with explicit time and speed modulation based on environmental semantics.

Circular Judgement and Sliding Window

Illustration of Circular Judgement in a 2D environment. For each waypoint, a circular region is defined. We compute a safety score based on obstacle proximity and trajectory continuity. If the score is below a threshold, we perform a sliding window optimization.

3D Optimization Animation

Optimization process of trajectory segments within the sliding window of the IKD-SWOpt framework.

Results on 3D and 2D Path

Illustration of the SHIFT Planner path optimization. The red dashed line represents the initial path, blue represents the optimized path, and green represents the smooth path.