𝗨𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 𝘁𝗵𝗲 𝗡𝗲𝘄 “𝗣𝗼𝗿𝘁𝗳𝗼𝗹𝗶𝗼 𝗔𝗰𝗾𝘂𝗶𝘀𝗶𝘁𝗶𝗼𝗻 𝗘𝘅𝗲𝗰𝘂𝘁𝗶𝘃𝗲” (𝗣𝗔𝗘) 𝗠𝗼𝗱𝗲𝗹: 𝗪𝗵𝗮𝘁 𝗜𝘁 𝗠𝗲𝗮𝗻𝘀 𝗳𝗼𝗿 𝗗𝗲𝗳𝗲𝗻𝘀𝗲 𝗧𝗲𝗰𝗵

The U.S. Department of Defense (DoD) is in the midst of a major overhaul of its acquisition system—and one of the centerpiece changes is the creation of the Portfolio Acquisition Executive (PAE).

This change signals a shift from traditional program-centric acquisition toward a portfolio-based, mission-driven model. Below is what’s publicly known so far.

🔍 𝗪𝗵𝗮𝘁 𝗶𝘀 𝗮 𝗣𝗔𝗘?

Under the upcoming reform, what were previously multiple Program Executive Officers (PEOs) managing discrete programs will increasingly become PAEs—senior officials responsible for a portfolio of related programs aligned by mission, domain, or technology.

A PAE’s portfolio could include several systems or capabilities (e.g., unmanned systems + autonomy + logistics + software), all under one accountable leader.

🧩 𝗞𝗲𝘆 𝗔𝘂𝘁𝗵𝗼𝗿𝗶𝘁𝗶𝗲𝘀 & 𝗥𝗲𝘀𝗽𝗼𝗻𝘀𝗶𝗯𝗶𝗹𝗶𝘁𝗶𝗲𝘀

They are expected to reallocate funds within the portfolio (e.g., divert funding from slower programs to higher-priority ones) to prioritize speed and relevance.
PAEs will be empowered to make trade-offs in favor of rapid fielding, rather than perfection. “Good enough, quickly” becomes a possible metric.
PAEs will serve longer stints (minimum ~4 years) and be evaluated on portfolio-level metrics such as delivery timelines, competition, and mission outcomes, rather than just compliance.


There are lots of open questions, in particular around budgets scope and scale and talent. Each service has been directed to submit portfolios within 60 days.

But one thing is clear: This reform is one of the most significant shifts in DoD acquisition culture in decades. 🚀

The first version of this post was published on LinkedIn in November 2025

Unresolved Software Challenges in Robotics in 2025

A previous post laid out the three key themes at this year's Bay Area Robotics Symposium (BARS) at Stanford.

In this post I will describe several open fundamental issues and some of the critical software challenges which were highlighted and remain unresolved:

𝟭. 𝗣𝗼𝗹𝗶𝗰𝘆 𝗥𝗲𝗹𝗶𝗮𝗯𝗶𝗹𝗶𝘁𝘆 𝗮𝗻𝗱 𝗦𝗮𝗳𝗲𝘁𝘆

Robotics still lacks robust mechanisms for real-time failure detection, safety guarantees under learned policies, and predictable behavior in out-of-distribution conditions


𝟮. 𝗘𝗻𝘃𝗶𝗿𝗼𝗻𝗺𝗲𝗻𝘁 𝗥𝘂𝗻𝘁𝗶𝗺𝗲 𝗕𝗼𝘁𝘁𝗹𝗲𝗻𝗲𝗰𝗸𝘀

Model inference is improving rapidly; simulation performance is not. Python-based environments remain a major constraint, and high-speed C++/GPU simulators are still nascent.


𝟯. 𝗧𝗵𝗲 𝗛𝘂𝗺𝗮𝗻–𝗥𝗼𝗯𝗼𝘁 𝗘𝗺𝗯𝗼𝗱𝗶𝗺𝗲𝗻𝘁 𝗚𝗮𝗽

Significant progress has been made, but mapping human intent onto robot morphology continues to be a major open challenge—especially for contact-rich or bimanual manipulation.


𝟰. 𝗟𝗼𝗻𝗴-𝗛𝗼𝗿𝗶𝘇𝗼𝗻, 𝗠𝘂𝗹𝘁𝗶-𝗦𝘁𝗲𝗽 𝗘𝘅𝗲𝗰𝘂𝘁𝗶𝗼𝗻

Memory remains a limiting factor. Retrieval-based methods represent progress, but long-sequence stability is unresolved for most architectures.


𝟱. 𝗨𝘀𝗮𝗯𝗶𝗹𝗶𝘁𝘆 𝗼𝗳 𝗥𝗼𝗯𝗼𝘁𝗶𝗰𝘀 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲

The ecosystem remains fragmented. Despite numerous advancements, there is no unified, developer-friendly stack equivalent to “the PyTorch of robotics.”


𝘛𝘩𝘦 𝘣𝘰𝘵𝘵𝘰𝘮 𝘭𝘪𝘯𝘦

Robotics is entering a period of accelerated capability—but progress is constrained less by hardware and more by software infrastructure, data engineering, and simulation bottlenecks.

👉 The largest opportunity now is to build the scalable, reliable software layer that bridges today’s innovations with real-world deployment at scale. 

This post was first published on LinkedIn in November 2025.