Exergi EPICC's energy dashboard. This screen showcases Exergi Predictive's value-add in densest form.
Overview
Exergi was chosen as the energy management arm of AAL's Robotic Combat vehicle (RCV) program.
Exergi Predictive wrote a gray-box AI/ML model to predict energy for off-road vehicles in contested environments. I ensured their capabilities were intuitive to operators.
Gray-box models balance explainability and speed. Typical AI models are "black boxes" meaning we're not sure how they derive their answer. Fully explainable models are slow. Exergi split the difference to ensure speed and explainability.
Contested environments are environments without dependable internet and other radio frequency connections. Edge machines enable the processing to happen on the vehicle.
A sales graphic. This drawing shows historical status (left) with Exergi Predictive's future statuses (right). Energy prediction is made possible thanks to their digital twin, simulation, and ML models.
The Challenge
Military energy prediction is an age-old problem. It’s even harder with autonomous robots.
Robotic combat vehicles—unmanned systems deployed in dynamic battlefields—depend on battery power to move, navigate, communicate, and use payloads.
Fluctuating energy demands from terrain, soil type, wind, mission type, and tempo lead to inefficiencies, reduced capabilities, or even mission failure.
The Army needed an AI-powered planning, prediction, and management tool for energy. From the operator's perspective, it also had to be intuitive, usable under pressure, and reliable in chaotic conditions.
An infographic that shows what Exergi owns (orange) and what it pulls from and contributes to (blue).
Process
The process was dynamic and iterative, reflecting the complexities of designing for emerging technologies.
Like many new projects, the process involved trial, error, and frequent adjustments. Insights accumulated gradually, pairing Exergi’s value with user needs.
How? Simply, through continuous learning and iteration. No design was sacred, we kept iterating. Because this was a new product, volatility and iteration were high.
Andrew and I met weekly, conducted four user visits, and engaged in ongoing collaboration with stakeholders and subject matter experts (SMEs).
Each design dead-end yielded valuable insights that shaped the final solution.
Here are the key human-centered activities we performed:
Stakeholder Collaboration
- Conducted in-depth discussions to define project goals, technical limitations, and opportunities.
- Maintained ongoing collaboration with SMEs and the tech team to align design with operational priorities.
Research
- Analyzed public and private resources to understand energy management in autonomous vehicles.
- Created proto-personas and journey maps to capture user needs and mission scenarios.
Information Architecture
- Developed IA diagrams and a sitemap to define the project’s scope and organize content effectively.
UI Design
- Designed a comprehensive UI framework, including widgets, graphs, and data visualizations.
- Iterated extensively to refine visual hierarchy and usability.
Custom Visual Assets
- Illustrated vehicles and designed non-standard icons to align with the project’s unique requirements.
- Illustrated various other infographics to communicate Exergi's value to customers.
Interaction Design
- Crafted interactions within and across widgets, sections, and screens to ensure seamless workflows.
Design Artifacts
- Finalized design elements, including typography, colors, components, and icons, to establish a cohesive visual identity.
Dev Feedback
- Critiqued the React prototype to align technical implementation with user needs.
One view possible on the map. User's can adjust view options and data layers to suit their current needs.
Solution
The human-centered process articulated, refined, and amplified Exergi's value-add across several military echelons.
This project was a nesting doll of projects. Some under one SOW, others under another. Some traditional product design, some strategic work. Here is a list of the key solutions we worked through:
- Mission Input & Route Selection: We refined the mission criteria and amplified what route has the highest probability of success.
- Energy Widget: It's hard to be more intuitive than a fuel gauge. In our case, we needed to display status for the battery and generator. We designed a dynamic widget that users can select how they think about energy: by time, by kWh, by distance. Users can expand that widget for more details on all aspects of energy.
- Energy Dashboard: Everything an operator needs to know about energy. This is the detailed energy screen.
- Prediction v Actual: Predictions are nice to see. Differences from the predictions are also nice to see. Eventually, a mission may deviate substantially from the predictions where predictions become noise. We gave the operator the ability to turn off predictions entirely.
- REx & Fan Controls: Most of the tool is 'view only', but does permit operators to control certain functions of the vehicle such as turning on/off the generator and fan. We ensured those controls always displayed their status. Expanding each item revealed their current status, event schedule, and a manual override toggle.
- Map, route, flags, & corridor: a dynamic map solutions showing energy optimal 'corridors', events as flags, and various layers such as 'known contested areas'.
- Timeline & Events: The timeline feature occurs throughout the tool and modifies it's shape to fit the context. We layered event flags onto the timeline to correlate to the map.
- Fuel Savings Optimization: Exergi allows users to optimize for fuel savings. This special setting adjusts the generator's parameters.
- Thermal Optimization for Stealth: RCVs are useful in stealth operations. The electric drivetrain allows them to be quiet, however, they also need to reduce their heat signature. Exergi offers a generator parameter to solve this, which also comes with a communication problem that we solved and embedded.
- Location Estimation: Remotely monitoring an RCV in a signal denied environment poses a unique challenge. Where is it? There is no signal to report back to the operator. Exergi created a more precise location estimation algorithm using energy and physics models. I designed a solution to ensure the operator knows the location is estimated and the range of confidence where the RCV could be.
Vehicle illustrations for Exergi EPICC. An aspect of the tool is inputting mission requirements so EPICC could simulate energy demands. We worked hard to make it easy to input mission requirements. These illustrations were a small step to make the experience more delightful.
Results
"You've increased our level of professionalism by a large step." — Will Northrop, Co-founder, Exergi Predictive
Andrew and I made a great design team deciphering needs, problems, and goals then applying Exergi’s value.
At the end, we uncovered a product strategy targeting a high value solution for a high value user. We humbly realized that Exergi's value-add is at a higher level thinker and planner.
This project was seen by other companies in the AAL cohort, one reached out for help on their tool. A nice vote of confidence.
Afterward
Reflections on Exergi
I really enjoyed this project — in fact, I learned that I really enjoy the energy domain. The project is ongoing, albeit less intense. It was a classic ambiguous request with a very short requirement list. Andrew and I shaped the product from the ground up.
From a user perspective, we listened to soldier input and made the necessary revisions. We also refined the product strategy seeking the right echelon user for maximal value add.
The project called for a nice mix of product design, product strategy, and marketing needs. Dynamic projects appeal to me as I acquire greater perspective on the product, company, and space they're competing in.
On a more detailed note, the project and process were riddled with challenges:
- Eastablishing User Clarity: Ensuring each screen provided a clear, actionable objective, avoiding information overload.
- Prioritizing and Visualizing Data: Sorting, grouping, and visualizing complex data into digestible, actionable insights for users.
- Balancing Autonomy and Manual Control: Designing an interface that showcased RCVs’ autonomous actions while allowing for seamless manual overrides. This included showing: 1) Current Actions: What's happening now. 2) Predicted Actions: What will happen
- Handling Predictions vs Actuals: Displaying historical data, real-time status, and future predictions without overwhelming users. Introducing a "difference component" to visualize the gap between actual outcomes and predictions.
- Integrating Energy Dynamics: Accounting for EREV drivetrain complexities, such as managing two energy sources (electric and generator) and displaying their combined status effectively.
- Various Macro Views: Depending on the context, operators need to see different "energy dashboards". We had to balance the experience between dashboards, maps with layers, timeline with layers, and mission settings.
- Various Units: Bridging user needs by converting energy into: 1) Distance: off-road range affect by terrain and surface factors. 2) Time: Operational endurance for stationary missions. 3) Traditional Metrics: Energy in kWh for technical understanding.
- System and Component Consumption: RCVs rely on a wide range of components with distinct energy demands. The drivetrain, onboard computer, weapon systems, radar, and sensors each consume varying amounts of energy. A primary challenge was providing operators with a detailed breakdown of energy consumption by component while presenting this information in a clear, actionable format. This required developing visualizations that allowed users to: 1) Identify high-energy-consuming components at a glance. 2) Monitor how individual systems, such as weapons or radar, impacted overall energy reserves.
The RCV program has changed hands in the government, so I'm unsure the long-term fate of our efforts. But, whoever discovers this work will have a big head start designing a solution in whatever application they need.
