CINEVE 2026 | Speech Review | Li Pengbo from Howso Technology: Human-Machine Collaboration Reshapes the Cabin Testing Paradigm, AI Breaks Through Three Major Industry Bottlenecks

The one-day AI-powered closed-door seminar on the automotive industry recently concluded successfully. At the event, Li Pengbo, CTO of Nanjing Howso Technology Co., Ltd., delivered a technical presentation titled "Efficient Human-Machine Collaboration Empowering Cabin Testing," which precisely addressed the pain points of the industry, sparking widespread resonance and attention within the sector.

Li Pengbo, CTO of Nanjing Howso Technology Co., Ltd., delivered a speech at the seminar.

1. Three Major Rigid Contradictions in Cabin Testing: The Traditional Model Can No Longer Sustain

In his presentation, Li Pengbo pointed out that as the intelligent cabin becomes the core difference between new energy vehicles and traditional fuel cars, its testing process faces three major rigid bottlenecks: the exponential increase in functional integration, the drastic compression of vehicle iteration and OTA cycles, and the requirement for 100% precision in safety and reliability. Traditional manual testing not only fails to handle the repetitive execution of tens of thousands of test cases, but also falls into a deadlock where “cost reduction, quality improvement, and efficiency enhancement cannot be achieved simultaneously.” Simply adjusting the labor scale cannot resolve the issue.

He emphasized that as the cabin becomes deeply integrated with intelligent driving, entertainment, and vehicle OS, new issues such as cross-system compatibility and physical touch failures are becoming more frequent. When test results are incorrect, it directly threatens the safety of the entire vehicle, a core contradiction that the industry must face.

2. AI Implementation in Cabin Testing: Adhering to the Bottom Line of Control and Traceability

Addressing common misconceptions in the industry, Li Pengbo clarified the technical principles: large models and intelligent agents are new forms of productivity, but the blind adoption of generic end-to-end solutions should be avoided. The core requirement of automotive testing is that the results must be unique. It is essential to avoid the illusions of large models and uncontrollable risks. All technological paths must be based on the premise of controllable processes, traceable stages, and verifiable results. This is the core difference between cabin AI testing and consumer-grade AI applications.

3. Engineering Path: Minimum Authorization + Process Decomposition to Achieve Specialized Division of Labor for Intelligent Agents

Li Pengbo shared Howso Technology’s practical implementation: relying on the Howso Xingqing Zhice platform, the testing process is atomized into stages such as requirement analysis, test case generation, task orchestration, and execution matching. Each stage is assigned a dedicated expert intelligent agent, strictly following the principle of minimum authorization—each intelligent agent is responsible for only one task. This approach reduces the risk of illusion errors from the source and ensures that the entire process remains controllable. It is the most suitable engineering solution for automotive testing at present.

Howso Xingqing Zhice Platform

4. Pure Software Testing Has Blind Spots, Physical Robots Restore Real-World Usage Scenarios

He specifically highlighted a pain point often overlooked by the industry: pure software testing cannot cover physical interaction defects, such as touch failure of screen-edge buttons, insufficient button sensitivity, and other issues where the underlying software shows no abnormalities but fails in actual use. To address this, the solution adopts a 6-axis robotic arm with a biomimetic hand to simulate manual touch, combined with visual, motion, and sound tri-validation to fill the gaps left by software testing and align with real-world usage scenarios.

5. Clear Division of Labor in Human-Machine Collaboration: AI Handles Repetition, Humans Guard Key Judgments

Regarding the division of labor in human-machine collaboration, Li Pengbo emphasized: "Our core strategy is 'AI handles repetition, humans guard the key.' AI is responsible for the rapid generation of tens of thousands of test cases, executing 24/7 without interruption, and determining simple visual outcomes such as light on/off and door/window opening; humans only intervene in key stages such as test point review and complex logic verification, for example, the frequency of ambient light flashing and multi-step functional interlinking effects. This ensures both efficiency and firmly upholds the 100% accuracy of the test results."

Nanjing Howso Technology Co., Ltd. Showcases Core Product Solutions, Including the Xingqing Zhice Cabin Robot, at the Exhibition, Attracting Professional Visitors from Home and Abroad

6. Li Pengbo Revealed Key Data on the Implementation of the Solution:

"Relying on this system, the efficiency of test case generation is improved by 10 times compared to manual methods, with daily output equivalent to the traditional testing volume of 3 days. Test execution efficiency has been enhanced by 4-6 times, perfectly adapting to the high-frequency regression testing requirements after OTA upgrades. More importantly, the entire testing process is traceable. It not only records the operation screen but also synchronously stores signal value change data, which significantly reduces the cost for developers to reproduce and fix bugs."

He also disclosed: "Currently, this solution has been deeply applied in mainstream automakers such as FAW Bestune, Geely, and Great Wall. In the field of cabin multimedia alone, the automation coverage rate has exceeded 50%. Three rounds of testing can achieve the same results as traditional manual full-volume testing, and the cost advantages of subsequent multi-round testing will become even more significant."

Future Plans: From Cabin-Specific to Full Vehicle Multi-Domain Collaborative Testing

Regarding future plans, Li Pengbo stated: "Nanjing Howso Technology Co., Ltd. will not stop at cabin testing. We are currently developing an exterior humanoid robot, and in the future, we will achieve 'in-vehicle robotic arms + exterior robots' for full-scene automated testing, covering all core testing stages inside and outside the vehicle. At the same time, we are working on 'digital immunity' products, extending the testing scope to multi-domain integration and collaboration across the entire vehicle, including the cabin, intelligent driving, chassis, and more. This will provide more comprehensive technical support for the automotive industry's intelligent transformation."