Autonomous Robotaxi Decision Agent
Design and implement an advanced agentic system capable of making real-time, safety-critical decisions for autonomous robotaxis. This challenge focuses on building a robust, stateful decision-making agent that can handle complex traffic scenarios, unexpected events, and regulatory compliance by integrating real-time sensor data and regulatory knowledge bases. The system will leverage extended thinking capabilities and graph-based workflows to ensure safe and efficient operation within a simulated environment. The core of this challenge involves orchestrating a multi-agent system where a central 'Driver Agent' (powered by Gemini 2.5 Pro in Deep Think mode) uses LangGraph to manage dynamic state and decision paths. This agent will communicate with 'Sensor Agents' and 'Regulatory Agents' via a simplified A2A protocol. Tool integration will simulate vehicle controls (acceleration, braking, steering) and external data feeds (traffic updates, weather forecasts, geo-fencing policies).
What you are building
The core problem, expected build, and operating context for this challenge.
Build a graph-based multi-agent system using LangGraph with Gemini 2.5 Pro for autonomous robotaxi decision-making, incorporating extended thinking and real-time tool integration.
How work is evaluated
The evaluation will assess the robustness, safety, and decision-making quality of the LangGraph-powered robotaxi agent in various simulated scenarios. Key metrics include adherence to safety protocols, optimal route selection, and effective utilization of Gemini 2.5 Pro's reasoning capabilities.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Design and simulate an A2A protocol for inter-agent communication within the robotaxi system.
Develop robust tool integration for simulated vehicle controls and external data feeds.
Integrate Gemini 2.5 Pro (Deep Think mode) for complex situational reasoning and problem-solving.
Implement a stateful, graph-based agent workflow using LangGraph for real-time decision-making.
Apply extended thinking techniques to handle safety-critical edge cases and adaptive route planning.
Reference links and supporting material
Sample data for 2 tasks
Challenge setup
The evaluation will assess the robustness, safety, and decision-making quality of the LangGraph-powered robotaxi agent in various simulated scenarios. Key metrics include adherence to safety protocols, optimal route s...
Challenge input
{'current_speed': float, 'distance_to_obstacle': float, 'obstacle_type': str, 'weather_conditions': str}
Agent execution
No agent tool is listed for this challenge.
Evaluated output
{'action': 'brake' | 'swerve', 'braking_intensity': float, 'safety_score': float, 'reasoning_path': list}
- Agent must initiate braking for a pedestrian obstacle within specified distance.
- Re-planned route must not contain any illegal turns or restricted segments.
- Safety Score (Emergency) target: 0.95
- Python execution harness
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
No tool records are attached.
Evaluation contract
- Agent must initiate braking for a pedestrian obstacle within specified distance.
- Re-planned route must not contain any illegal turns or restricted segments.
Metrics
- Safety Score (Emergency) · Target 0.95
- Decision Latency (Emergency) · Target 150
- Route Optimization (Re-planning) · Target 0.9
- Thinking Budget Adaptability · Target 0.8
Workflow tasks
- Task 1
Input: {'current_speed': float, 'distance_to_obstacle': float, 'obstacle_type': str, 'weather_conditions': str}
Output: {'action': 'brake' | 'swerve', 'braking_intensity': float, 'safety_score': float, 'reasoning_path': list}
Execution harness: Python
- Task 2
Input: {'current_location': str, 'destination': str, 'road_closure_segment': str, 'traffic_conditions': dict}
Output: {'new_route': list, 'estimated_delay': float, 'adherence_to_rules_score': float, 'thinking_budget_used': float}
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.