Multimodal Physics and Sensor Simulation for Synthesizing Training Data for Robot Manipulation
Multimodal Physics and Sensor Simulation for Synthesizing Training Data for Robot Manipulation

© Fraunhofer IPA

© Fraunhofer IPA

ICRA 2023 Virtual Manipulation Challenge: Stacking

Context

Vanderlande is a logistic solution provider for the Airports, Parcel and Warehousing segment. The increasing demand for automation from our customers is pushing us to challenge the limits of state-of-the-art robotic solutions. In today’s logistic operations there is still a lot of heavy manual work in the loading processes. The technical complexity of automated loading solutions is high, but the poor working conditions are increasing the urgency for those solutions. Through this challenge, Vanderlande wants to raise awareness for this topic in the robotics community and scout partners to co-develop robotic loading solutions.

The Challenge

The goal of this challenge is to achieve a dense and stable packaging of objects inside a container. 

The challenge takes place in a virtual environment (extended Bullet environment with ROS integration) and focuses on the 3D bin packing problem. The motion planning and control of the robot arm are out of scope for this challenge.

As a participant you are asked to develop a software component that outputs place poses for a set of objects that have to be loaded in a container. As input you will receive the dimensions of the objects and the container, and as output we expect a pose for the object to be loaded. The stacking environments will provide point clouds of the carrier after each packed object.

The challenge will assume top loading with a suction cup gripper, such that objects may only be rotated around the z-axis.

The Use Cases

The track is split into a parcel loading and item packing task, which are described below.

Parcel Loading

© Vanderlande B.V.

For the parcel loading task, a set of reference parcels will be given in form of dimensions and weights. The set is limited to carton boxes with rigid cuboid shapes and is sampled from the distributions below.

The carrier is a K. Hartwall Big Box with internal dimensions of 2141 x 1076 x 990 mm (see left image below). The goal of the parcel loading task is to stack the incoming boxes in a compact and stable way to utilize the space in the box as efficiently as possible (see right image below).

The parcel loading task consists of 3 scenarios:

  • Adhoc packing (no lookahead, no sequencing)
  • Hybrid packing with lookahead (lookahead=20, no sequencing)
  • Hybrid packing with lookahead and sequencing (lookahead=20, sequencing buffer=3)

A lookahead of 20 means that at every placement, the weight and dimensions of the next 20 parcels are known. A sequencing buffer of 3 means that at every placement the order of the next 3 parcels can be changed.

Item Packing

For the item packing task, a set of objects representative for a warehousing customer in the food segment will have to be packed into a 600 x 400 x 270 mm carrier (see left image below).

The set will include complex, rigid shapes. The goal of the item packing task is to stack the incoming items in a compact and stable way to utilize the space in the box as efficiently as possible (see right image above). The scenario in the item packing track will consider adhoc 3D bin packing. Only the very next product will be known, and it has to be placed directly into the carrier.

 

Participating teams can opt to participate solely in either one or in both of these stacking tasks.

Evaluation

Objects will be spawned at the suggested pose and the Bullet simulation will be used to evaluate their motion within the carrier. These motions will be used to calculate static stability scores for each element in the sequence. After the whole object sequence is placed, small oscillatory motions will be applied to the carrier to measure the overall stability of the final stacking result. Together with the volumetric fill rate (= sum of volumes of packed packages / volume of container) and planning time, those are the criteria that will be used to evaluate the packing efficiency.

The table below shows the weights that will be used to score the different criteria.

CRITERIA C WEIGHT W
Space utilization (volumetric fill rate) 8
Static stability 7
Planning time 3
Stability under stress (bin vibrations) 1

 

The task will be evaluated for a fixed number of packing sequences. The total score is aggregated as the sum of the results scored over all tested sequences.

Organizers

Tim Nickel, Reem Al-Gaifi 

Award

The Stacking Award for the »Parcel Loading« task is sponsored by Vanderlande Industries B.V.

There is a cash prize for the 3 contestants with the highest score:

1st prize: € 6.000

2nd prize: € 3.000

3rd prize: € 1.000

Furthermore, one contestant will be selected to jointly explore product development of a parcel loading solution.

 

There is no dedicated award for the »Item Packing« task, participation is for the fame of winning ;-)

The Virtual Manipulation Challenge is organized by the research project Sim4Dexterity. The Sim4Dexterity project has received funding from the German Federal Ministry of Education and Research (BMBF) through grant 01IS21061 (Sim4Dexterity).