As robotics continues to move from controlled environments into more complex, real-world settings, the role of research institutions in shaping the next generation of systems is becoming increasingly significant.
One of the organizations operating at this intersection is Mitsubishi Electric Research Laboratories (MERL), the North America-based R&D arm of Mitsubishi Electric.
Led by president and CEO Anthony Vetro, MERL focuses on advancing core technologies that underpin robotics, including perception, control systems, and machine learning. Its work spans multiple domains, from industrial automation to energy systems, with an emphasis on bridging the gap between theoretical research and practical deployment.
That gap remains one of the central challenges in robotics today. As Vetro explains in this interview, systems that perform well in controlled lab environments often struggle when exposed to the variability and unpredictability of real-world industrial settings. Addressing this requires not only improvements in sensing and control, but also a deeper integration of physical understanding into AI models.
The conversation highlights several areas of active development, including more precise robotic manipulation and force control, predictive sensing to anticipate human and object movement, and new approaches to robot training using augmented reality and audio-visual interfaces. These technologies, Vetro suggests, could lower deployment barriers and enable robots to adapt more quickly to new tasks.
The discussion also touches on the growing concept of “physical AI” – systems capable of operating within the constraints of the physical world – and where meaningful progress is already being made.
While fully autonomous operation in unstructured, human environments remains a work in progress, the direction of travel is clear: robotics is becoming less about isolated machines and more about integrated, adaptive systems designed for real-world complexity.
Interview with Anthony Vetro
Robotics & Automation News: Many robotics breakthroughs originate in research labs but struggle to reach commercial deployment. What do you see as the biggest barriers to translating advanced robotics research into real-world industrial systems?
Anthony Vetro: A major barrier is the gap between controlled lab settings and the complexity of real-world industrial environments. Robots can often perform well in the structured settings they are developed in yet struggle when they encounter variability and human interaction on the factory floor.
Another challenge is integration into existing workflows where reliability and cost efficiency are critical. Real progress depends on systems that better reflect the physics of the real world to create robotics that are able to operate fluently outside the lab.
R&AN: MERL is working on improving robotic manipulation and force control. How close are we to robots being able to handle delicate, contact-rich tasks with the same reliability as humans?
AV: There has been steady progress in terms of manipulation and force control, but we are still not at human-level reliability in contact-rich tasks.
Humans adapt instantly to sensory feedback and changing conditions, where robots still struggle to generalize outside the data they are trained on. Improvements in sensing and control are closing that gap, creating tighter integration of perception with physics-based reasoning so robots can respond more naturally in real-time.
R&AN: You mentioned integrating sensing to anticipate human and object movement. How important is predictive capability for enabling robots to operate safely alongside people, and what are the key technical challenges still to overcome?
AV: a. Predictive capability is essential for robots that share space with people. Once a system is able to anticipate human motion and object behavior, it can act safely and efficiently in dynamic environments.
The key challenge is dealing with uncertainty in real-world settings where behavior is not always predictable. Stronger models that combine sensing with an understanding of physical dynamics will be important for enabling smooth collaboration in factories and other shared workspaces.
R&AN: Training robots has traditionally been time-consuming and data-intensive. How do approaches such as augmented reality and audio-visual interfaces change the economics and scalability of robot training?
AV: Augmented reality and audio-visual interfaces allow robot training to become more scalable and accessible across various industries and locations. Instead of only relying on traditional programming, operators are able to guide robots directly in context, and with a reduced setup time. It also helps robots learn from demonstrations and feedback in a more natural way. These approaches lower the barrier to deployment and allow robots to adapt faster to new tasks and environments.
R&AN: How closely aligned are the priorities of research labs like MERL with the immediate needs of industry, particularly in sectors such as manufacturing, logistics, and healthcare?
AV: There is strong alignment between MERL’s research and industry priorities in manufacturing, logistics, and healthcare. Our focus remains on foundational technologies that work to support real-world deployments.
Our work connects closely to practical challenges like automation, efficiency, and safety. The goal is to develop capabilities that can transition from research into systems that operate reliably at scale.
R&AN: There is growing discussion around “physical AI” systems that combine perception, decision-making, and action in the real world. How does MERL view this concept, and where do you see the most meaningful progress being made today?
AV: Physical AI is about systems that understand and operate within the laws of the physical world. As models become more grounded in physics, they can deliver better performance and lower costs in complex, real-world environments.
For example, one area seeing meaningful progress today is data center energy efficiency, where physical AI can dynamically control airflow and direct cooling where it’s needed most, reducing energy use while enabling more proactive maintenance.
R&AN: Looking ahead, what breakthroughs or milestones would signal that robots are truly ready to operate reliably in unstructured, human environments at scale?
AV: Key milestones would include consistent performance across new environments and safe interaction with humans in shared spaces. Robots will need to be able to respond reliably under uncertainty and changing conditions. When systems are able to combine perception, reasoning, and physical understanding in real time, we will be much closer to true deployment at scale in unstructured environments.
