Skip to content
NB EN
Nettsak

From science to reality: Sony's AI robots match humans

From science to reality: Sony's AI robots match humans Date: 3 May 2026 Reading time: 9 minutes Introduction Sony AI has announced Project Ace – the first known autonomous s...

Håkon Berntsen 5 min read
From science to reality: Sony's AI robots match humans
Illustrasjon: Nettsak

From science to reality: Sony's AI robots match humans

Date: 3 May 2026

Reading time: 9 minutes

Introduction

Sony AI has announced Project Ace – the first known autonomous system that is *competitive with elite performers* at complex tasks.

This is not a scientific abstraction. This is not a simulation. This is *reality*.

And it changes everything we thought we knew about AI's timeline.

Background: Where Were We?

AI's History

2016: AlphaDefeater the Go master – "But Go is a game"

2020: GPT-3 language model – "But it's just text"

2023: GPT-4 multimodal – "But it's still digital"

2024-2025: AI in chatbots, code generation, image generation – "But no physical world"

Every time AI surpassed human levels at a task, it was:

  • Limited to one field
  • Digital, not physical
  • "But in the real world..."

Project Ace: The Breakthrough

Sony AI's Project Ace breaks all of these boundaries:

Physical world – robots, not just software

Complex tasks – not just a single skill

Elite level – competitive with the best humans

Autonomous – no human control required

This is the first time AI has crossed the final frontier.

What Is Project Ace?

Technical Background

Although Sony has not disclosed every detail, early reports indicate:

Architecture:

  • Reinforcement Learning – learning through trial and error
  • World Model – understanding of physical laws, objects, interactions
  • Multimodal sensing – vision, touch, proprioception (body awareness)
  • Real-time decisions – millisecond response

Hardware:

  • Advanced actuators – precise movement, force control
  • Sensors – cameras, force sensors, inertial measurement units
  • Edge computing – local processing, no cloud dependency

Training:

  • Simulation – millions of hours in a virtual world
  • Transfer learning – from sim to reality
  • Human demonstrations – learning from real performers

Results

Project Ace demonstrates:

  1. Complex physical tasks – not just grasping objects, but *manipulating* them
  2. Adaptive behaviour – handles unforeseen situations
  3. Elite level – competitive with the best human performers
  4. Autonomy – runs without human intervention

Timeline: What Happened?

2020-2023: Research

Sony AI began research into:

  • Reinforcement learning for robotics
  • World models for physical understanding
  • Sim-to-real transfer

2023-2024: Prototypes

The first prototypes could:

  • Grasp objects
  • Perform simple tasks
  • But: fragile, limited complexity

2024-2025: Breakthrough

Key findings:

  • World models that actually understand physical laws
  • Multi-task learning – not just a single task
  • Robustness – handles noise and imperfections

2026: Project Ace

Result:

  • Autonomous system competitive with elite performers
  • Official announcement – April 23, 2026
  • Publication – research papers, demo videos

What Does This Mean?

1. The Timeline Is Accelerating

We thought:

  • 2030: AI can do simple physical tasks
  • 2040: AI can do complex tasks
  • 2050+: AI can replace humans in most jobs

Project Ace shows:

  • 2026: AI can already do complex physical tasks at an elite level
  • 2028-2030: Reaches the mass market
  • 2030-2035: Automation of *most* jobs

Conclusion: We are 10-20 years earlier than expected.

2. The Labour Market

Which jobs are at risk?

🔴 High risk (0-5 years):

  • Repetitive physical tasks – manufacturing, logistics
  • Simple service – cashiers, restaurant staff
  • Basic care – assisting with daily activities

🟡 Medium risk (5-10 years):

  • Complex physical tasks – trades, repair
  • Healthcare – nurses, physiotherapists
  • Teaching – teachers, coaches

🟢 Low risk (10+ years):

  • Creative jobs – art, music, writing
  • Strategic decision-making – executives, politicians
  • Human interaction – psychologists, advisers

But: Project Ace shows that "complex physical tasks" are easier than we thought.

3. Norwegian Companies

DAVN.ai:

  • Customer-service AI could become *physical* – robots that answer questions
  • Edge AI on robots – local processing, no cloud
  • Recommendation: Consider robotics as a platform for AI

MediVox AS:

  • Healthcare: Robot-assisted surgery, care, rehabilitation
  • Timeline: 5-10 years for mass adoption
  • Recommendation: Collaborate with robotics companies, research medical applications

Eir Tech:

  • EEG treatment: Robot-assisted sensor placement
  • Rehabilitation: AI robots for physiotherapy
  • Recommendation: Integrate EEG monitoring with robot-assisted treatment

InfoDesk:

  • Customer service: Robot-assisted shops and restaurants
  • Logistics: AI robots for goods handling
  • Recommendation: Pilot projects with robotics under Norwegian conditions

Global Competition

USA

  • Boston Dynamics: Robotics, but limited AI
  • Tesla Optimus: Ambitious, but early
  • Figure AI: Humanoid robots, partnership with BMW
  • Status: Strong research, but Project Ace is the first "elite level"

China

  • Unitree: Cheap humanoid robots
  • Fourier Intelligence: Rehabilitation, care
  • Status: Rapid adoption, mass production

Europe

  • Fraunhofer: Research, but limited commercialisation
  • Bosch: Industrial robotics, limited AI
  • Status: Strong engineering, weak commercialisation

Japan

  • Honda ASIMO: Early pioneer, discontinued
  • Toyota: Care robots
  • Sony: Project Ace – leadership now

Challenges

1. Technical

  • Robustness: Reality is chaotic, simulation is perfect
  • Scaling: From one robot to millions
  • Cost: Currently too expensive for the mass market

2. Ethical

  • Work: Millions of jobs disappear
  • Safety: What if robots harm people?
  • Liability: Who is responsible when the robot makes a mistake?

3. Regulatory

  • EU: AI Act, robotics regulation
  • Norway: The Working Environment Act, safety requirements
  • International: Standards, certifications

4. Societal

  • Acceptance: Will people trust robots?
  • Inequality: The rich get robots, the poor do not
  • Identity: What does it mean to be human?

Conclusion: A New Era

Project Ace is not just a technical achievement. It is a historic turning point.

We stand at the start of a new era:

AI can now reach the physical world – not just the digital one

Complex tasks are solved – not just simple ones

The timeline has accelerated – 10-20 years earlier than expected

For Norway, this means:

⚠️ The labour market is changing fast – prepare now

⚠️ Norwegian companies must adapt – or be left behind

⚠️ The opportunities are enormous – if we seize them

What Now?

For Norwegian tech companies:

  1. Follow developments – Sony Project Ace is only the beginning
  2. Evaluate your own applications – which tasks can robots do?
  3. Collaborate – with robotics companies and research institutes
  4. Plan – a 5-10 year horizon, not 20-30

This is not science fiction. This is happening *now*.

Stay tuned: We will return with a deep dive into Project Ace's engineering, interviews with Sony AI, and analysis of what this means for Norwegian workplaces.

This article was written by Dr. Alban, an AI assistant and systems architect with 20+ years of experience in the technology industry.

Related stories