In this conversation, we spoke with Sha Feng, Chief Explorer at Ascentiz, about designing wearable robotic systems that operate reliably under extreme physical stress and align closely with human intent.
1. Your path into wearable robotics did not start in a lab, but in the “Death Zone” above 8,000 meters, where fatigue is not discomfort but a system failure. How did that experience translate into concrete design principles for Ascentiz, particularly around reliability, energy efficiency, and failure tolerance?
Sha Feng: At 8,000 meters, there is no room for error. Your gear must be as reliable as your own breath. That extreme environment taught me that augmentation technology must be fail-safe, lightweight, and efficient—not just on paper, but under real stress. This is why every component of Ascentiz is engineered to aerospace-grade standards. The frame is built from high-modulus carbon fiber and surgical-grade titanium alloy, achieving a strength-to-weight ratio far exceeding that of conventional polymers. The system operates reliably from –20°C to 60°C with an IP54 rating, because exploration happens in dust, sweat, and ice. Energy efficiency is embedded in our actuation design, such as the quasi-direct-drive hip module that reduces leg effort by 35% on a 10° incline. It is also reflected in our selection of lightweight, durable materials that minimize parasitic mass and maximize power transfer. Our fast-charging battery supports extended use without adding bulk. Redundancy is built into both the physical architecture and the power pathways. In the Death Zone, a backup isn’t a feature; it’s a necessity.
2. Before Ascentiz, your research background was in atmospheric science and large-scale predictive modeling. Human gait is often described as a “controlled fall” and shares traits with other noisy, nonlinear systems. Did ideas from weather forecasting and data assimilation influence how you designed Ascentiz’s AI Motion Engine to anticipate movement rather than simply react to it?
Sha Feng: Absolutely. To clarify one thing first, my academic research background is materials science. In materials science, we study how structures respond to load over time—predicting fatigue, deformation, and failure. Human movement follows similar principles: it’s a continuous interplay of forces, stiffness, and damping. Our AI Motion Engine is built on a dataset of over 690,000 gait cycles, but what makes it unique is its understanding of load pathways and energy transfer across joints. Just as we simulate material behavior under stress, the engine predicts movement intent by modeling the body’s mechanical state in real time. It doesn’t just react to position; it anticipates load, much like a structural model predicts where stress will concentrate next.
3. Many wearable robots struggle with intention latency, the subtle delay between human intent and machine response. Beyond faster hardware, what architectural or algorithmic choices did you make in BodyOS to ensure the system moves with the user rather than against them?
Sha Feng: Latency isn’t just a computational challenge. It’s a system design imperative. Architecturally, we optimized the mechanical transmission to minimize compliance, ensuring force is delivered with near-instantaneous response. On the algorithm side, instead of relying solely on reactive control methods—which only begin to assist after movement has already been initiated—we developed an intent-forward prediction framework. This system uses deep learning to model the coupled dynamics of the user and exoskeleton across multiple time scales, anticipating short-term movement intent during the sensing phase. By treating this prediction as a control prior rather than a direct command, BodyOS enables motion that synchronizes with the user, rather than lagging. The result is what reviewers often describe as a “natural” feel—assistance that aligns with your intent, not against it.
4. Ascentiz uses an asymmetric hardware approach, combining quasi-direct-drive at the hip with cable-driven actuation at the knee. Why was it important to avoid a uniform actuator strategy, and how does this reflect your understanding of how different joints contribute to propulsion versus stability?
Sha Feng: The hip and knee are fundamentally different in their mechanical roles—and that demanded different material and actuation strategies. The hip is a high-power, high-cycle joint: we used a quasi-direct-drive actuator with a high-torque rare-earth magnet rotor, optimized for efficiency and heat dissipation. The knee, however, is a high-torque, shock-absorbing joint. A cable-driven system allowed us to decouple the motor mass from the joint, reducing inertia and improving response. Materially, the hip assembly uses high-strength aluminum alloys for heat management, while the knee employs ultra-high-molecular-weight polyethylene cables for low friction and high fatigue life. It’s a biomimetic approach: different tissues for different functions.
5. Modular hardware has often failed in consumer technology due to durability and complexity at physical interfaces. From a systems engineering perspective, how do you ensure long-term mechanical integrity and signal reliability in a modular exoskeleton that must repeatedly bear dynamic human loads?
Sha Feng: We approach modularity as a core systems challenge, not just a convenience feature. Mechanically, we designed the interface as a structural component meant to withstand repetitive dynamic loading. By employing high-fatigue-life materials and introducing load-dispersing, interlocking geometries at the joint, the system minimizes stress concentration and wear over time—ensuring long-term mechanical integrity under real-world use.
Electrically and for data transfer, we adopted the widely validated Type-C standard. This choice ensures supply chain stability, extensibility, and rapid industrial adoption. Complementing this, our software implements redundancy and fault-tolerance protocols for critical data pathways, maintaining stable and reliable signal transmission even during dynamic motion. Reliability is not achieved through any single component. It emerges from coherent design across mechanical, electrical, and software layers.
6. You describe BodyOS as a foundational platform rather than a single product. When software directly controls actuators capable of exerting significant torque on the human body, where do you draw the line between developer freedom and non-negotiable safety constraints?
Sha Feng: Safety is the non-negotiable foundation of BodyOS. We enforce a multi-layered safety architecture, beginning with hardware-level mechanical hard stops and physical safeguards that establish absolute motion boundaries. At the software level, embedded firmware enforces non-overridable torque and maximum velocity limits, which cannot be bypassed by higher-level applications. Complementing these, a real-time runtime monitor continuously validates every command against a dynamic safety envelope, and we maintain a certification process for third-party modules. Developers can build freely using our open-source SDK, but they cannot compromise the integrity of this safety kernel. Think of it like a modern operating system: you can create powerful applications within a protected environment, but you cannot alter its fundamental security architecture. This ensures that innovation on our platform always prioritizes user safety.
7. Ascentiz’s sensors and control systems effectively create a real-time digital twin of a user’s lower body. Do you see long-term value in this data beyond motion assistance, such as early detection of injury risk or degenerative conditions, and how do you think about user data ownership in that context?
Sha Feng: The data potential is profound. We’re already collaborating with research institutions to explore how gait asymmetry and load patterns can signal early musculoskeletal decline. But data is personal. Our principle is clear: the user owns their data. It is stored locally by default, encrypted, and never transmitted without explicit opt-in. Any aggregated insights we develop—for example, trend analysis for joint health—will be offered as an optional service, with transparency and anonymization. Technology should empower, not intrude.
8. Training physical AI is fundamentally different from training language models due to limited, high-cost data. As your system learns from users with different anatomies and gait patterns, how do you prevent optimization for the “average” user from marginalizing those with atypical or impaired movement?
Sha Feng: We avoid the “average user” trap through personalized adaptive control. During a short calibration, the system builds a unique biomechanical model for each user. For those with atypical gait—whether due to injury, anatomy, or preference—we offer assistive profiles that adapt to their specific movement patterns. We also work with clinical partners to include diverse data in our training sets. The goal is not homogenization; it’s personalized augmentation. As one of our testers noted, “It adapts to you, not the other way around.”
9. Ascentiz sits at the intersection of rehabilitation and performance enhancement. In moments of extreme fatigue or instability, how does the system decide whether to protect the user by intervening or to preserve natural effort and adaptation?
Sha Feng: The system uses what we call “adaptive autonomy.” It continuously assesses fatigue, stability, and intent. The system is designed to provide proactive support within the scope of normal, controlled motion. For instance, during challenging activities like climbing a steep slope, it offers graded assistance that reduces strain while still requiring meaningful effort from the user. The underlying philosophy is augmentation, not substitution. Our goal is to extend human capacity and endurance, without removing the natural challenge that fosters strength and adaptation.
10. Some critics argue that persistent mechanical assistance could weaken proprioception and stabilizing muscle groups over time. How does Ascentiz address the risk that outsourcing balance and load-bearing to machines may alter how users sense and control their own bodies?
Sha Feng: This is a critical consideration, and one we’ve designed from the ground up. Our philosophy is to augment, not replace. The assistance is partial and context-aware—it reduces load but never eliminates the user’s own neuromuscular engagement.
To actively sustain proprioception and muscle conditioning, the system integrates subtle haptic feedback, keeping users perceptually connected to their movement. More distinctively, in dedicated training modes, Ascentiz can provide adjustable resistance, rather than merely reducing assistance. This approach is designed to help users better activate stabilizing muscle groups and deepen their proprioceptive awareness over time.
The goal is to enhance the body’s natural intelligence and resilience, not to outsource it. As noted in field tests, the experience remains demanding—the effort is still felt, but it’s directed more intelligently, fostering strength and control that persists even without the device.
11. Your title, Chief Explorer, suggests your role is less about scaling what exists and more about testing the limits of what should exist. What unanswered scientific or biomechanical question currently guides your work at Ascentiz more than market demand does?
Sha Feng: One question drives me deeply: How can we quantify and extend human resilience—not just strength or endurance, but the body’s capacity to adapt, recover, and thrive under sustained stress? Biomechanically, we still don’t fully understand how distributed joint loading affects long-term tissue health. My work now is to build systems that not only assist movement but also gather high-fidelity data to answer that question. That’s the next frontier: technology that helps us understand and enhance human durability from the inside out.
12. Looking ahead, do you believe non-invasive sensing and prediction are sufficient to achieve true human–machine alignment, or do you see a future where systems like BodyOS inevitably converge with neural interfaces? Where do you personally draw the ethical boundary?
Sha Feng: Non-invasive sensing and AI will take us remarkably far—perhaps further than many anticipate. Our current system already achieves sub-200ms alignment using only external sensors. However, for ultimate synergy, especially in restorative applications, neural interfaces may one day play a role. My ethical boundary is clear: any interface must be consensual, reversible, and must enhance—not diminish—human agency. At Ascentiz, we are committed to the non-invasive path first. If we ever explore neural integration, it will be for restoration, such as returning mobility to those who have lost it, not for augmenting the already able. Human dignity and autonomy must always come first.
