Sort waste using local recycling rules
Gemini Robotics-ER 1.5 looks up local recycling rules online, then plans and sorts objects into compost, recycling and trash bins. Task: Suggested implementation: Use Gemini Robotics-ER 1.5 as a high-level reasoning layer for a robot. The workflow involves querying Google Search for local recycling rules, identifying visible objects, and generating a spatially grounded sorting plan. The model then calls a vision-language-action (VLA) model or user-defined functions to execute physical movements. Human role: Developers must integrate the model with robot hardware APIs and ensure safety compliance.; tools: Gemini. Human role: The original post does not say how a person stayed involved.; limitations: Model-level safety filters do not replace physical robot safety engineering. Developers are responsible for implementing emergency stops, collision avoidance, and risk assessments. The model is in preview, implying potential API changes or stability issues. Integration requires custom code to connect reasoning outputs with specific robot hardware or VLA models..
Tools
Published on Agent Work Atlas:
Overview
Gemini Robotics-ER 1.5 looks up local recycling rules online, then plans and sorts objects into compost, recycling and trash bins.
Workflow
Suggested implementation: Use Gemini Robotics-ER 1.5 as a high-level reasoning layer for a robot. The workflow involves querying Google Search for local recycling rules, identifying visible objects, and generating a spatially grounded sorting plan. The model then calls a vision-language-action (VLA) model or user-defined functions to execute physical movements. Human role: Developers must integrate the model with robot hardware APIs and ensure safety compliance.
Reported result
The source demonstrates a rendered animation of a spatially grounded sorting plan with interleaved text and points. It does not provide end-to-end success rates for physical sorting tasks. The model is currently in preview, and actual performance in real-world deployment is not quantified in the provided text.
Human role
The original post does not say how a person stayed involved.
Limitations
Model-level safety filters do not replace physical robot safety engineering. Developers are responsible for implementing emergency stops, collision avoidance, and risk assessments. The model is in preview, implying potential API changes or stability issues. Integration requires custom code to connect reasoning outputs with specific robot hardware or VLA models.
Try it
Act as a robotics developer integrating Gemini Robotics-ER 1.5. I need to prototype a waste sorting agent. 1. **Context**: Target users are facility managers testing automated sorting. Output location: Local JSON file and console log. 2. **Inputs**: [replaceable: Image URL of a table with mixed waste items], [replaceable: City name for recycling rules]. 3. **Steps**: - Use the Gemini API to call Google Search for recycling guidelines in the specified city. - Analyze the provided image to identify objects and their spatial coordinates (2D points). - Generate a sorting plan mapping each object to a bin (compost, recycling, trash) based on the rules. - Output the plan as a JSON object containing object labels, bin assignments, and normalized [y, x] coordinates. 4. **Safeguards**: Before executing any physical commands, print the plan for human review. Do not send commands to a robot. 5. **Quality Check**: Verify that all objects in the image are accounted for and that bin assignments match the retrieved local rules. 6. **Error Handling**: If an object is ambiguous, flag it for human decision. If search fails, use default general rules and note the fallback. Start with a dry run using a sample image and city. Ask clarifying questions if the image format or API keys are missing.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.