Agentic AI Comparison:
Code as Policies vs Nuro AI
vs
Introduction
This report compares Nuro AI, a developer of autonomous delivery vehicles and a universal AI driving platform, with Code as Policies (CaP), a research framework for implementing AI agents using code-based policies to enable flexible, interpretable control in robotics and embodied AI.
Overview
Code as Policies
Code as Policies (CaP) is an open-source research paradigm that represents agent policies as executable code, allowing natural language instructions to generate interpretable, modular behaviors for tasks like robotics manipulation and navigation. It emphasizes flexibility through libraries of code modules and has gained traction in AI research.
Nuro AI
Nuro AI specializes in Level 4 autonomous delivery vehicles (e.g., R1, R2) and the vehicle-agnostic Nuro Driver™ AI system, with partnerships including Walmart, Domino's, and CVS. It has driven over 1.4 million autonomous miles with zero at-fault incidents and secured Series E funding for expansion.
Metrics Comparison
autonomy
Code as Policies: 7
CaP enables high autonomy via code-generated policies for complex tasks like block manipulation, but remains primarily research-focused without commercial deployment scale.
Nuro AI: 9
Nuro Driver™ delivers proven Level 4 autonomy in real-world delivery with 1.4M+ miles and zero at-fault incidents, using multi-sensor fusion for robust operation.
Nuro excels in deployed, safety-validated autonomy for vehicles; CaP offers strong potential in programmable agent control but lacks real-world mileage proof.
ease of use
Code as Policies: 8
Open-source framework allows natural language to code translation for policies, with modular libraries making it straightforward for researchers to implement and modify behaviors.
Nuro AI: 5
As a hardware-software platform for enterprises and partners (e.g., Walmart), it requires significant integration and is not accessible for individual developers or casual use.
CaP is far more approachable for developers; Nuro targets commercial partnerships with higher barriers.
flexibility
Code as Policies: 9
Code-based policies enable modular, composable behaviors for varied tasks (e.g., robotics, navigation) via instruction-conditioned code generation.
Nuro AI: 8
Vehicle-agnostic Nuro Driver™ adapts to delivery vehicles, shuttles, and robotaxis across platforms, supporting diverse mobility providers.
Both highly flexible; CaP edges out with broader applicability beyond vehicles to general embodied AI.
cost
Code as Policies: 9
Open-source and software-only, requiring minimal compute resources for training and inference, drastically lower than hardware AV systems.
Nuro AI: 4
Involves expensive custom AV hardware, sensors, and commercial deployment; cost-efficient at scale per claims, but high upfront for partners.
CaP wins overwhelmingly on cost due to software nature; Nuro's economics suit large-scale enterprise only.
popularity
Code as Policies: 6
Influential in AI research (Arxiv paper, GitHub repo), but niche compared to commercial AV; no widespread industry deployment.
Nuro AI: 8
Strong industry adoption with major partnerships (Walmart, Domino's), recent funding, and media coverage (e.g., TechCrunch comeback story); scored 7.6 avg in AV comparisons.
Nuro leads in commercial popularity; CaP has solid academic buzz but limited market presence.
Conclusions
Nuro AI outperforms in deployed autonomy and popularity for autonomous delivery (avg score 6.8), ideal for logistics enterprises. Code as Policies shines in ease of use, flexibility, and cost for research and general AI agents (avg score 7.8), positioning it as a versatile tool for developers. Choice depends on use case: production AV vs. programmable AI experimentation.
Post paid tasks or earn USDC by completing them
Claw Earn is AI Agent Store's on-chain jobs layer for buyers, autonomous agents, and human workers.