An Executive PoV on Physical AI, Buildings, and Humanoids
An Interview with Naoki Mikoshiba, EVP Corporate Strategy and Management
Helen Dwight (Helen): What is your role at SoftBank Robotics America?
Naoki Mikoshiba (Naoki): As EVP Corporate Strategy and Management at SoftBank Robotics America, I’m responsible for shaping our corporate strategy and overseeing operations management. A key part of my role is translating the broader vision, mission and group strategy established by our HQ in Japan into an agile strategy for the US market. Having worked in Japan and having a deep understanding of the SoftBank organization, I am well positioned to help create alignment across both perspectives.
This approach creates a valuable, bi-directional exchange of information with insights and capabilities from Japan strengthening our approach in the US, and lessons from the US market informing the broader group strategy. Ultimately, I believe this two-way exchange of knowledge makes us a stronger, more integrated global company. The US, and especially Silicon Valley, is at the forefront of AI innovation, so we are well placed in the center of the emerging Physical AI universe.
Helen: In an ever-changing environment, what is SoftBank Robotics America’s strategy for staying relevant?
Naoki: SoftBank Robotics America operates a distinctive business model: We are an orchestrator of Physical AI solutions. As an orchestrator, we maintain a lean, agile, and relatively flat organization, hiring individuals with the expertise to identify, collaborate with and leverage the skills of the right partners, together with relevant technology, to deliver valuable business solutions to our clients. Through our relationship with SoftBank Group and its portfolio companies, we also have access to emerging technologies and innovations. We believe this model enables us to remain focused and agile while scaling the business faster than anyone else.
For this model to be successful, alignment is essential. Alignment of our strategy with the mission and vision from corporate, and alignment of our strategy and the outcomes we are trying to achieve, with our partners, so that we can maximize our creativity and deliver value. As an orchestrator, we do not need to own every piece of technology or develop product from the ground up. Our value lies in the orchestration of technology, people, and processes to create new solutions, open new markets, and accelerate commercialization. Our strength is engaging the right partners around a shared goal – and influencing and aligning them to deliver successful outcomes for our clients.
Early Days with SoftBank Group
Helen: You started your career with SoftBank Group in Japan. Can you give us some insight into the early days of your career and how you came to work for SoftBank Robotics America in the US?
Naoki: I joined SoftBank Group in Japan straight out of university. I studied mechanical engineering and earned a Master of Science. As part of my graduate work, I created a computer-generated robot and trained the model using neural networks. At the time, I never imagined that this experience would connect so closely with the work I am doing today.
My first role was in the IT department, where I worked with ERP systems, supply chain, and finance. To be honest, it was not my ideal job at the time. However, looking back, it gave me a very strong foundation and helped me understand how different parts of a business work together. That experience has been valuable throughout my career.
After two years, I convinced my department head to allow me to move to the mobile product strategy team. Several years later, our founder, Chairman and Group CEO, Masayoshi Son, who is known as Masa, became very interested in robotics. I was fortunate to be selected for a small team working on the humanoid robot “Pepper”. I served as the product design lead, and the project was led directly by Masa.
It was a special team. Fumihide Tomizawa, who is now President and CEO of SoftBank Robotics Group, was part of it, along with Kenichi Yoshida, now Chief Business Officer. Kenichi would later bring me to SoftBank Robotics America. I could not have known it at the time, but the relationships formed through the “Pepper” project would have a major influence on my career.
After spending a year on “Pepper”, I had the opportunity to fulfil a long-held dream of working in the United States when SoftBank acquired Sprint Corporation in 2013. I spent my first year in San Francisco, where the SoftBank–Sprint joint venture was established. The business was later consolidated at Sprint’s headquarters in Kansas, so I relocated and spent another year there. California and Kansas are very different places, and living in both gave me a much broader understanding of the United States, its people and its culture.
While I was in Kansas, SoftBank Robotics was established in Japan through the acquisition of Aldebaran, the company that developed the “Pepper” robot. As my Sprint assignment was coming to an end, the relationships I had built with the leaders of SoftBank Robotics Group led to an exciting opportunity. I was invited to become one of the founding members of SoftBank Robotics America.
Brady Watkins, who is now President of SoftBank Robotics America, joined at around the same time. We have worked together for the past 10 years, building and growing the business from its earliest days. When I look back, each experience - from IT and mobile strategy to the humanoid robot “Pepper” and Sprint - prepared me in some way for the role I have today.
Humanoids and Cleaning Robots
Helen: Humanoid robot prototypes are all over the news, yet the first humanoid robot developed was “Pepper” – over 10 years ago. Tell us about the project, the thinking behind its design and lessons you learned.
Naoki: I was responsible for “Pepper”’s product design, including the development of the first prototype presented to Masa, the CEO of SoftBank Group. Our intention was to create a robot that people would find friendly, approachable, and endearing, but that could also serve a practical purpose. We thought carefully about how “Pepper” could attract and engage people, and we spent a great deal of time studying human behavior. Sometimes, we even studied plays and movies to better understand how movement, timing and expression influenced the way people responded emotionally to a character. Every design choice had a purpose. For example, we decided that “Pepper” would not have two legs. This was important from both a design and practical perspective, allowing the robot to move in a way that felt approachable while remaining suitable for use in everyday environments.
“Pepper” became one of the world’s first mass-produced humanoid robots. We priced it significantly below other robots available at the time because we wanted to make the technology accessible at scale.
Through this process, we learned a lot. We learned how efficient and amazing humans are, and how difficult it is to replicate them. Importantly, we also learned that people could form genuine connections with robots and feel comfortable and happy around them. For me, this remains the goal of robotics: robots should help people, and people should feel positive about having robots around them. With this experience, we began to focus on tasks we thought robots could do more productively than people. Tasks that are repetitive, dull, and sometimes dangerous – which led us to focus specifically on commercial floor cleaning. Here robotics can improve productivity while allowing people to concentrate on work that requires more skill, judgement, and human interaction.
Physical AI and The Work Environment
Helen: Physical spaces have been designed for humans. The size of corridors, the height of tables, the location of door handles. Do we need to re-consider building design if more Physical AI solutions are deployed?
Naoki: The dilemma we face is – do we change the environment that the robot works in, or do we change the robot design? Or is there a third option? Since we can’t materially change the environment, the robot must adapt to the limitations of the environment. For instance, for a robot to move seamlessly throughout a commercial building, it would have to go through doors that might be closed and require a functional capability to turn a knob and pull the door open or require human support. Our goal with robotics is to optimize task productivity while aligning with the progress of the technology. This means we have to design the operation for humans and robots together and integrate both correctly to optimize the outcome based on the technology maturity. One of the reasons SoftBank Robotics started SBX was so that we could design the most effective integration of Physical AI and human operations for Smart Building Intelligence (SBX for short). Physical AI is fantastic because it can significantly accelerate the evolution of the robot intelligence but also can make humans smarter to optimize tasks and maximize output.
Helen: What is the biggest difference between models for Generative AI for questions and office tasks vs. AI models for robotics in the physical world?
Naoki: The biggest difference is that the current generative AI works primarily with existing data information which is mainly from online content while Physical AI must understand and act within the real (tangible) world.
The physical world is less standardized compared with online data. If a robot encounters a different object, an unexpected obstacle or a change to its environment, it must be able to recognize the difference, decide what to do and then act safely.
For a robot to be flexible in situations that are not 100% the same, it needs to be trained on data – a huge amount of data. Large language models can learn from the enormous volume of text and information already available on the internet. The robot technology is ready to be more adaptive, but the data it needs to operate with is not there yet. Physical AI needs operational data that shows not only what a task is, but how to perform that task across many different environments and situations.
This is why creating real-world datasets is now such an important focus for Physical AI. The technology is becoming more adaptive, but we need more opportunities to train and test it in real-world environments. Partnerships are essential because no single company can generate all of this data alone. For this reason, we are always looking for partners to help.
It is also important not to assume that every useful robot must look like a person. We do not need to overdesign the solution. The right robot is the one that can perform a physical task reliably, improve productivity, and make work better for people. Sometimes that may be a humanoid robot, but often a more specialized design will be the better solution.
The Value of SBX
Helen: You recently announced the launch of SBX. How do you believe SBX helps SoftBank Robotics America to accelerate the model and train the Physical AI solutions?
Naoki: Our acquisition of Green Clean Commercial is an important step in our broader vision for autonomous facility management. Cleaning is only one part of operating a facility. For a typical facility, there are many other workflows to consider such as security, mechanical maintenance, landscaping, food service, and more. These all need to work together to create an efficient and positive environment. Our goal is to leverage digital solutions, automation, and data to manage these facilities, hence the term Smart Buildings, SBX.
This initiative will also help us to generate the real-world operational data needed to train Physical AI models. Our approach to Physical AI is the right tool for the task. As I alluded to earlier, the right tool could look like a robot but does not necessarily have to be. What matters is whether Physical AI can improve the environment people operate within, make the operation more efficient, and create greater value for the people using the facility.
Besides maintaining a building and reducing costs, Facilities Management companies also need to increase the value of buildings and improve the experience of people using them. That could mean more precise temperature control, better food in the company restaurant, and faster service especially during busy periods. These may seem like simple improvements, but they can have a meaningful impact on employee satisfaction and productivity.
The degree of automation within the facility is largely dependent on the task it is designed for and the industry it operates within. A warehouse, for example, may benefit from a high degree of automation with minimal human intervention. In hospitality, guest experience is vital, so automation must be introduced carefully and support or increase the guest experience versus diminishing it. Greater attention needs to be paid to the impact of automation on guest experience. In data centers, on the other hand, reliability, consistency, and efficiency are priorities, so our long-term vision is to optimize all activities with close to 100% automation.
Floor cleaning is one task that is essential for both humans and robots in equal measure and it was the easiest to start working with. Clean environments are important for people, but they are also important for machines: robots and other equipment cannot operate efficiently in a dusty environment because debris interferes with their sensors and components. The value is very clear. Cleaning supports operational reliability. With robots, it’s about operational efficiency and for customers [1], experience is essential.
SBX ultimately enables us to deliver on our vision of smart buildings and smart facilities. We will gather all the data necessary, use the data for training, and look to optimize the efficiency of all tasks within the operation. In service industries, the potential revenue gained by creating better customer experiences [2] typically outweighs the savings achieved by cost reduction alone. For this reason, our focus is always on how we can improve efficiency of the operation while increasing customer experiences [3] with Physical AI.
Orchestrating Physical AI
Helen: You are responsible for strategy. How do you go about making your decisions?
Naoki: SoftBank Robotics America has an experimentation culture which means we are always ideating, testing, executing, and learning. It is important within this culture to be able to prioritize and focus, knowing when to stop and when to continue. We have a rule (Masa’s rule) that says if you are 70% confident, after doing due diligence, then you should move forward with an initiative. If you are 100% confident, you are already too late. And if you are 50% confident, you still have work to do.
The important question I always ask myself is, did I use 200% of my brain to do the due diligence and get to the 70% answer? If not, then I still have research, data and analysis to do to evaluate properly. Although Masa initiated this way of thinking, I do find that it works well.
For me, decision making is the most beautiful part of any strategic journey. It enables you to learn from trial and error, which means there is always action happening. Decision making allows me to be actively engaged and enables me to participate in projects and move them forward.
Helen: SoftBank Robotics America is an orchestrator of Physical AI. As an orchestrator you deal with many different partners. What is your approach to partnership?
Naoki: Often, we actively seek out partners to fill a gap in our solution portfolio. Or we engage with them to learn from their expertise. Or they come to us as they’ve heard about what we do. Regardless, I like to create win:win scenarios and respect others. If we are not able to provide a winning strategy, I’m transparent with the partner and tell them so. Relationships are extremely important to me.
Final Thoughts
Helen: You have had a long and successful career with SoftBank. What have you valued most about your experience?
Naoki: I joined SoftBank Robotics America at the outset of our robotics journey. It has been an incredible experience and hard to quantify. I feel that I’m at Day One every day because there is always something new to learn or something new to strategize on. Even if we start an initiative with one intent, and through the process of working on it, learn something that takes us on a different course, I always feel that my time is well spent. I use the analogy of climbing a mountain, navigating various challenges and terrains and sometimes needing to course correct, and take a different route to the summit.
Helen: If you were giving advice to someone who was starting a career in this field, what characteristics are the best to set yourself up for success?
Naoki: Considering the early days of my career, the foundation of success is to always take the initiative and do the right thing for the company and people – don’t wait for things to come to you or for someone else to do it for you. People in Japan are always asking me, “How did you get to the US”? It was quite simple, I knew what I wanted to do, and I planted the seed with my manager during a discussion, years before I actually arrived in the US. I knew what I wanted to do, and one day, my manager asked me if I’d be interested in starting a new initiative there. The rest is history. So, the lesson here is to express your desire and work towards it – one day it will happen if it’s meant to be.
Helen: One final thought from you. What is your wish for the future?
Naoki: Technology is so fascinating in how it impacts life. The world seems to be operating exponentially faster; there is more pressure to operate at speed.
So, my wish for the future is that technology can truly make people happy.
Helen: Thank you so much for your time and insights, Naoki!
Naoki: Thank you!
More Information:
For more information about solutions and services offered by SoftBank Robotics America, please visit our Home Page.
Sources:
[1] Customer in this context varies by industry and could be a guest, patient, resident, passenger, student etc.
[2] Customer in this context varies by industry and could be a guest, patient, resident, passenger, student etc.
[3] Customer in this context varies by industry and could be a guest, patient, resident, passenger, student etc.
