Foresight Packing Challenge API
Product · API / developer platform · Engineering & development by Dexterity
Build and test autonomous packing agents with production-grade Physical AI logic.
A REST API for building and testing autonomous packing agents in a simulated physical environment. It allows developers to solve high-density spatial reasoning challenges using Dexterity's physics-based world model logic.
- Price
- From 0 USD
- Languages
- English
Category: Developer Tools
- APIs and SDKs
- REST API with JSON support
- Platforms
- API
- Limits
- 50 compete games per day; max 2 concurrent games per API key
- Deployment
- Cloud-based API service
- Languages
- English (documentation)
- Truck Depth (X-axis)
- 2.0 m
- REST API for autonomous packing agents
- Full physics simulation in Compete Mode
- Supports 3D position and quaternion placements
- Real-time density scoring and state updates
Overview
The Foresight Packing Challenge API is a REST-based developer platform designed for researchers and students to build and test autonomous packing agents. It provides a simulated environment for solving complex spatial reasoning problems, specifically maximizing density and stability in a constrained truck volume. Developers can use Dev Mode for rapid prototyping without physics constraints or Compete Mode for realistic evaluations where boxes settle based on physical properties. The API handles 3D placements using positions and quaternions and provides real-time feedback on packing density. This platform allows the community to interface with a specialized component of Dexterity's world model used in production logistics.
API and Integration
- REST API Integration: The REST API allows developers to send box placements and receive post-placement game states, including physics-settled positions.
- JSON Data Format: Developers can interact with the API using JSON-formatted requests and responses for all game actions.
- Game State Management: The API supports starting games, placing boxes with 3D positions and quaternions, and tracking remaining boxes.
- Standardized Base URL: The platform provides a base URL at dexterity.ai/challenge/api for all service requests.
Development and Testing Modes
- Dev Mode Prototyping: Dev Mode provides no physics simulation, where boxes land exactly where they are placed for rapid prototyping.
- Compete Mode Simulation: Compete Mode features full physics simulation where boxes settle realistically to test production-grade logic.
- Concurrent Game Support: The API allows up to 2 in-progress games per API key to support concurrent testing.
- Daily Compete Limits: Developers can play up to 50 games per day in Compete Mode per API key.
Packing Logic and Evaluation
- Density Scoring Logic: The system calculates packing density as the fraction of usable truck volume filled with boxes.
- Gap Penalization: The scoring formula penalizes gaps behind the pack front by using the deepest box face in the denominator.
- Statistical Ranking: The platform supports ranking based on the interquartile mean (IQM) of an agent's most recent 50 compete games.
Use cases
- Researchers use the API to test new spatial reasoning algorithms for high-density truck loading.
- Students build automated agents to compete in packing challenges by optimizing box placements in a 3D volume.
- Developers prototype packing logic in Dev Mode to verify placement coordinates before running full physics simulations.
Specifications
- APIs and SDKs
- REST API with JSON support
- Platforms
- API
- Limits
- 50 compete games per day; max 2 concurrent games per API key
- Deployment
- Cloud-based API service
- Languages
- English (documentation)
- Truck Depth (X-axis)
- 2.0 m
- Truck Width (Y-axis)
- 2.6 m
- Truck Height (Z-axis)
- 2.75 m
- Max Concurrent Games
- 2 per API key
- Compete Mode Rate Limit
- 50 games per day
- Data Format
- JSON
- Authentication Method
- API Key in request body
- Minimum Games for Ranking
- 25 games
How it works
- Start a Game Session: Users start a game by sending a POST request to the start endpoint with their API key and chosen mode (dev or compete).
- Place Boxes via API: The agent receives box dimensions and weight, then sends a placement request containing the box ID, 3D position, and orientation.
- Receive State Updates: The API returns the post-placement state, including the settled positions of all boxes and the number of boxes remaining.
Pricing
- API Access
- Free for eligible.edu users
FAQ
What is the difference between Dev Mode and Compete Mode?
Dev Mode has no physics simulation or rate limiting, while Compete Mode uses full physics simulation and is rate-limited to 50 games per day.
How many concurrent games can I run?
The API supports up to 2 in-progress games concurrently per API key.
How is the packing score calculated?
Scores are based on packing density, calculated as the total volume of placed boxes divided by the volume defined by the deepest box face and truck cross-section.
What are the dimensions of the truck?
The API uses a standard truck segment with a depth of 2.0m, width of 2.6m, and height of 2.75m.
Added by AntSpace from public sources (dexterity.ai).