I’ve been getting this question a lot lately. Based on what I’ve observed in the industry and shifts in the hiring market, here are some practical thoughts.
The Bottom Line: Good Prospects, but Sharp Divergence
Embodied AI and reinforcement learning are not “uniformly good” fields. The demand varies significantly across sub-domains, company types, and roles.
One-sentence summary: algorithm roles are hyper-competitive, engineering roles are in short supply, and application roles are just taking off.
Algorithm Roles: Competitive, but High Ceiling
RL algorithm engineer is one of the hottest directions, but the competition is fierce.
Demand side:
- Big-tech AI Labs (DeepMind, OpenAI, Meta, ByteDance, Tencent): continuously hiring, but the bar is extremely high — top-tier conference papers plus a PhD from a prestigious school are basically required.
- Humanoid robotics companies (Unitree, Zhiyuan, Fourier, etc.): aggressively expanded hiring in 2024–2025, but started tightening in 2026 with a stronger emphasis on real-world deployment ability.
- Autonomous driving companies (Waymo, XPeng, Huawei): the planning and decision-making modules need RL backgrounds; demand is stable.
Salary ranges (2026, China, for reference only):
- Fresh PhD graduate: roughly ¥400K–800K/year (big-tech SP offers can reach ¥1M+, but spots are very limited)
- 3–5 years of experience: roughly ¥600K–1.2M/year (varies widely by company, region, and individual ability)
- Senior / expert: ¥1.5M+ (these positions are extremely scarce)
The reality: headcount for algorithm roles is shrinking. During the 2024 humanoid robotics boom, companies hired heavily. By 2025–2026, many found that real-world deployment was slower than expected and began optimizing their teams. People with pure algorithm skills and no engineering capability face the greatest risk.
Engineering Roles: Talent Shortage, but Demanding a Hybrid Background
This is the direction I think is undervalued.
Embodied AI is not just an algorithm problem — it’s an engineering problem. Deploying models onto real robots requires solving:
- Real-time control: policy inference latency must stay under 10 ms
- Hardware integration: sensor calibration, actuator drivers, communication protocols
- System optimization: CUDA acceleration, model compression, edge deployment
- Simulation platforms: building and maintaining large-scale parallel simulation environments
People who can do all of this are in very short supply. And these roles have relatively relaxed education requirements — a bachelor’s degree plus hands-on experience is often sufficient; a PhD is not mandatory.
Salary ranges (for reference only):
- 3–5 years of experience: roughly ¥500K–900K/year (varies widely by company, region, and individual ability)
- Senior engineer: roughly ¥800K–1.5M/year (senior positions are limited)
Not much lower than algorithm roles, but far less competition.
Application Roles: Just Getting Started, Biggest Opportunity
This is the direction I’m most bullish on.
The technology behind embodied AI has moved past the “can we do it” phase and entered the “how do we use it” phase. Factories need to integrate AI capabilities into existing production lines, logistics companies need robots for sorting, hospitals need surgical assistance — these scenarios don’t require you to invent new algorithms; they require you to apply existing technology to concrete problems.
Typical roles:
- Robotics Application Engineer: responsible for task adaptation and deployment in specific scenarios
- Sim-to-Real Engineer: transferring policies trained in simulation onto real robots
- Technical Solutions Engineer: interfacing with customer requirements and designing technical solutions
Demand for these roles is growing rapidly, but supply is insufficient. Traditional robotics engineers don’t know AI, AI engineers don’t know robotics, and very few people understand both.
Salary ranges (for reference only):
- 3–5 years of experience: roughly ¥400K–700K/year (varies widely by company, region, and individual ability)
- Senior / management: roughly ¥700K–1.2M/year (these positions are relatively scarce)
Starting salaries may not match algorithm roles, but the growth potential is greater. And because these roles are close to the business, they’re less likely to be cut during optimizations.
A Few Trend Predictions
- Pure algorithm roles will keep getting more competitive; engineering and application roles will appreciate in value.
As foundation models (large language models, foundational world models) become increasingly powerful, the value of “hyperparameter tuners” is declining. The value of people who can actually deploy models is rising.
- Cross-disciplinary talent is the most sought-after.
People who understand AI + robotics + a specific industry (manufacturing, logistics, healthcare) will be in high demand for the next 3–5 years.
- Startup opportunities are at the application layer.
Foundation models are a big-tech game, but the application layer has a huge number of niche scenarios that remain uncovered. For example, “using world models for rapid task adaptation in factory robots” — this is a direction that big companies overlook, but the market demand is very real.
- Geographic concentration is pronounced.
Beijing, Shanghai, Shenzhen, and Hangzhou account for over 80% of all positions. Opportunities in other cities are scarce unless there’s a local robotics industry cluster (e.g., Dongguan, Suzhou).
Advice for People with Different Backgrounds
Students: If you have the opportunity to pursue a PhD, do it. The bar for algorithm roles will only keep rising. But during your PhD, make sure to do work with engineering value — not purely theoretical research.
Career changers: Don’t go straight for algorithm roles — the competition is too intense. Enter through engineering or application roles, build practical experience, then pivot toward algorithm work if desired.
Traditional robotics engineers: Learn PyTorch, learn RL fundamentals, learn simulation tools (MuJoCo, Isaac Sim). Your existing robotics knowledge is an advantage — add AI on top and you’ll be highly competitive.
Pure software engineers: Learn robotics fundamentals, study control theory, and get familiar with hardware. Your programming skills are an advantage — add the physical-world piece.
A Final Thought
The prospects for embodied AI and reinforcement learning are good, but that “good” is not evenly distributed. It belongs to those willing to deeply understand problems and get their hands dirty solving real ones — not to those just chasing the hype.
My own focus is on world models + Sim-to-Real, sitting at the intersection of engineering and application. I don’t have big-tech resources or an academic pedigree, but by consistently producing content and accumulating hands-on experience, I’m gradually building my own influence.
It’s not a fast road, but once you get through it, it’s solid.
Comments