The robotics industry has largely focused on building machines that perform tasks efficiently, but a new wave of embodied AI is asking a fundamentally different question: what if robots were designed to build relationships instead? OlloBot, a Hong Kong-based company, is showcasing its OlloNi SS1 companion robot at IFA Berlin 2026 this week, marking a significant pivot in how the physical AI industry thinks about human-robot interaction.

What's Driving the Shift From Task-Focused to Emotionally Intelligent Robots?

While most of the embodied AI industry remains focused on service robots, home assistants, and humanoid robots built for task efficiency, OlloBot is taking a different path. The company positions its OlloNi SS1 within a new category it calls "Cyber Pet," which sits at the intersection of emotional AI and physical AI. Rather than optimizing for what robots can do, OlloBot is prioritizing what relationships they can build.

"The key question is not only what robots can do, but what relationships they can build," said Lyn Fang, founder of OlloBot.

Lyn Fang, Founder, OlloBot

This philosophy reflects a broader recognition within the robotics research community that embodied AI systems face distinct challenges depending on their purpose. Dr. Chia-Yen Lee, Senior Editor of IEEE Transactions on Automation Science and Engineering, notes that autonomous vehicles, embodied AI humanoids, and home robots all share foundational technologies like computer vision and spatial mapping, but they operate under vastly different physical constraints, behavior models, and safety thresholds.

How Are Researchers Addressing the "Small Data" Problem in Physical AI?

One of the most pressing challenges in advancing embodied AI is what researchers call the "small data" problem. For emerging product categories like humanoid robots and companion robots, physical production volumes remain low, which means there simply aren't enough real-world interactions and test cycles to generate the large volumes of training data that traditional machine learning models require.

Dr. Lee identifies several research directions that could help overcome this bottleneck and accelerate physical AI development:

• Physics-Informed Neural Networks (PINNs): These models combine physical kinematic equations with empirical manufacturing metrics to bridge the gap between sparse real-world data and robust AI training, allowing researchers to generate synthetic data that reflects actual physical constraints.
• Dynamic Sensor Fusion: Modern robots combine optical cameras, tactile arrays, inertial measurement units, and strain gauges. Using dynamic latent-space modeling and continuous-time temporal alignment can reduce hundreds of asynchronous sensor streams into low-dimensional representations while preserving critical performance characteristics.
• Sim-to-Real Calibration Loops: The behavior of embodied AI models depends heavily on how closely a factory's digital twin reflects actual physical assembly variances. Closed-loop feedback systems between shop-floor sensors and AI training pipelines can continuously update simulation parameters based on real plant distributions.

The OlloNi SS1 demonstrates how these principles might translate into consumer products. The robot is designed to interpret user interactions and contextual signals, enabling more natural and empathetic engagement through its EmpathCore emotional intelligence engine, Affinilog interaction system, and Memory Heart Module, which preserve meaningful memories and foster relationships that evolve over time.

Steps for Researchers to Bridge Theory and Real-World Physical AI

• Ground Research in Industry Collaboration: Seek out research initiatives that directly address industrial shop-floor bottlenecks. Clean theoretical datasets rarely capture the messy, sparse, and asynchronous realities of production environments, so working closely with industry partners exposes researchers to genuine edge cases that transform theoretical algorithms into production-grade solutions.
• Cultivate Interdisciplinary Networks: Connect early and consistently with professional societies like IEEE RAS (Robotics and Automation Society), open-source communities, and cross-disciplinary peers. Complex engineering domains require diverse expertise spanning mechanical hardware, advanced materials, computer science, and information management.
• Pair Physics with Data-Driven Architecture: Combine physics-informed and explainable AI with modern distributed architectures, understanding how to deploy lightweight, low-latency inference at the edge while orchestrating heavy model training and digital-twin governance in the cloud.

Dr. Lee emphasizes that resilient breakthroughs often emerge at the intersection of domain physics and data science. He notes that "the quality of your networks defines the quality of your academic life," highlighting how professional engagement and collaboration shape research trajectories in embodied AI.

Dr. Lee

What Does the Emotional AI Category Mean for the Broader Robotics Industry?

OlloBot's positioning of the OlloNi SS1 as a companion robot rather than a task-oriented machine signals a maturation in how the industry thinks about embodied AI. While the company originally planned a Kickstarter crowdfunding campaign for late August, it has rescheduled the launch to reflect its commitment to long-term product readiness and a thoughtful rollout strategy amid evolving regulatory and market conditions.

The shift toward emotional AI and companionship represents a recognition that not all physical AI applications require humanoid form factors or task efficiency. Instead, some of the most meaningful applications of embodied AI may lie in creating machines that understand evolving user preferences and develop meaningful relationships through continuous companionship and shared experiences. This approach challenges the industry's prevailing assumption that robots must be optimized for productivity and instead asks whether they might be optimized for presence and connection.

As the field of physical AI continues to mature, the research challenges identified by experts like Dr. Lee, combined with commercial innovations like OlloBot's emotional intelligence framework, suggest that the next generation of embodied AI will be defined not just by what robots can do, but by the quality of the relationships they can sustain.