PackingEnv simulates sequential placement of three-dimensional items into trailers. It turns physical packing constraints into a reinforcement-learning environment configurable with item types, container shapes, and an item sequence. section 3.4
An episode connects physical placement decisions to the agent’s feedback:
- The observation contains each trailer’s 3D occupancy grid and information about the current item. section 3.4
- An action selects a container, a position, and one of six item orientations; placement checks enforce container bounds and prevent overlap. section 3.4
- Rewards encourage packing items, while penalties discourage invalid placements, invalid actions, skipped items, and excessive episode steps. section 3.4
Random-agent sanity check
A random policy tries to pack 12 items across four differently sized trailers; the resulting 3D view makes placement outcomes visible. section 3.4
Figure 3.7 Visualization of one instance of trailer loading and packing environment.
View a rendered example of items packed into trailers of different shapes. section 3.4
