[Context and motivaton] Advancements in machine learning (ML) have impacted driving automation systems (DAS) as well as ML-based perception systems. This increased complexity leads to intensified multi-party collaboration and special needs for requirements representations.[Question(s)/Aim(s)] Well-defined requirements are essential in this safety-oriented domain, but requirements engineering (RE) for ML-enabled perception systems remains challenging. We aim to contribute to the cross-section of requirements representations, multi-party collaboration, and the need for a shared language for ML-enabled automotive perception systems development in DAS. [Method] A case study with ten experts from a major automotive original equipment manufacturer (OEM), its suppliers, and researchers is conducted, followed by a thematic analysis.[Principal idea(s)/Results] Current practices for requirements representations in this context rely on natural language, operational design domains (ODDs), and key performance indicators (KPIs). Multi-party collaboration uses formal communication, while RE for ML-enabled systems faces challenges from a lack of mature standards and causality issues.[Contribution/Conclusion] We identify previously unrecognized practical limitations in requirements representation for ML-based perception systems and multi-party collaboration. We highlight effective practices, and our findings suggest that a shared language and reference architecture could address key RE challenges.