← Course Home Module 0 · Start Here: The Two Worlds
Module 0 · Foundations

Start Here: The Two Worlds

Before any math or machine learning, you need to understand the two real-world problems the papers tackle. This module assumes zero background in energy systems or robotics — it is written for a smart newcomer.

▶
Audio recap
A ~2-minute spoken summary of this module — great for revision on the go.
Why this module exists An oral exam almost always opens with: "In your own words, what problem does this paper solve, and why does it matter?" If you can't answer that in plain language, nothing technical that follows will save you. This module gives you that answer for both papers.

0.1 World 1: Solar power and the grid (Paper 1)

Solar farms are power plants made of panels

A utility-scale photovoltaic (PV) plant is a field of solar panels feeding electricity into the national grid. The one in Paper 1 is huge: rated at 75 megawatts (MW), with about 312,500 individual PV modules covering roughly a square kilometre in South Africa's sunny Northern Cape. The panels' DC power flows into 84 inverters — devices that convert DC to grid-ready AC — and then to the grid connection point.

The problem: the sun is not reliable

The grid must balance supply and demand every second. But PV output is intermittent: a passing cloud can slash a plant's output within minutes. Grid operators therefore distrust solar unless they can see it coming. That is power forecasting: predicting the plant's output ahead of time so operators can plan reserves, balance the grid, and trade energy. Paper 1 targets intra-day forecasts, 1 to 6 hours ahead, at 15-minute resolution — 21 future values predicted at once — using only on-site historical measurements (no external weather forecasts).

An analogy Think of the grid operator as a chef cooking for a banquet where guests appear and vanish unpredictably. A forecast is the guest list for the next few hours: even an imperfect one transforms panic into planning.

Macro vs micro: the paper's actual question

Nearly all published forecasting models are trained on the plant's total output — a macro-level model that treats the whole farm as one big panel. But a large plant is not uniform: wind cools some rows more than others, clouds and terrain shade parts of the field, so different segments behave differently — up to 3% output variation between inverters just from wind direction. Paper 1 asks: would 84 small "low-level" forecasts (one per inverter), added together, beat one big macro forecast? And: can standard deep learning models capture that low-level behaviour at all?

What Paper 1 does Using four years of real plant data, it trains three standard deep learning models — a feedforward network (FFNN) and two recurrent networks, LSTM and GRU (Modules 3 and 5) — at both macro level and inverter level. To make 84 inverter models affordable it invents an inverter-clustering trick (Module 8), and it wraps everything in a scrupulously fair 3-phase hyperparameter search. The headline finding is refreshingly honest: aggregated inverter-level forecasts improve accuracy only marginally (best NRMSE 8.02% vs 8.12% macro) — so macro-level models are good enough, which is genuinely useful news for practitioners.

0.2 World 2: Robots that model themselves (Paper 2)

Every robot is born in simulation

Before a robot arm is built, engineers create a CAD model (a precise 3D blueprint) and derive its kinematics — the equations linking each motor's joint angle to where the arm's parts end up in space. With those equations a robot can plan: "to put my fingertip there, set my joints to these angles." This works, but it is hand-built, rigid, and wrong the moment the robot is damaged, worn, or modified.

The dream: a robot that learns its own body

A child learns their body not from blueprints but by watching themselves move — in a mirror, for instance. Paper 2 gives a robot the same ability: the robot waves its arm around randomly (motor babbling) in front of a single ordinary 2D camera, and from those video frames alone — no 3D scanners, no measuring tape, no kinematic equations — it learns a self-model: an internal, queryable simulation of its own shape and movement.

The mirror test, for machines Only a few animals — humans, chimpanzees, orangutans — recognize themselves in a mirror. The paper frames its work as a small step in that direction: a robot using vision of itself to build and update an internal model of its own body, including noticing when its body has changed.

How the self-model works, in one breath

The learned model — the free-form kinematic self-model (FFKSM) — answers a simple query: "Given my current joint angles, is the 3D point (x, y, z) occupied by my body, and would the camera see it?" Ask that question for thousands of points and you can render the robot's predicted silhouette, plan collision-free motions, and control the arm — all through the model. It is inspired by NeRF, a computer-vision technique for learning 3D scenes from 2D images (Module 6). Training compares the model's rendered silhouette to the actual camera image; nothing else is needed.

What Paper 2 does With 12,000 video-frame/joint-angle pairs from motor babbling, a 4-degree-of-freedom arm learns its FFKSM. The model then: (1) predicts the robot's 3D morphology more accurately than baselines; (2) performs motion planning — tracing a 3D spiral, dodging obstacles — without any hand-written kinematic equations, using gradient descent through the model itself; and (3) detects damage (a bent link) and recovers by fine-tuning on fresh post-damage video. The whole model fits in 333 kB.

0.3 Why one course covers both papers

A solar farm and a robot arm look unrelated, but the exam machinery underneath is largely shared:

Shared conceptPaper 1 (Solar)Paper 2 (Robot)
Core taskRegression: predict future power valuesRegression: predict occupancy/visibility of 3D points
Input dataSensor time series (power, irradiance, weather)Video frames + joint angles
Neural architecturesFFNN, LSTM, GRUCoordinate/kinematic encoders + predictive module (fully connected, ReLU)
Loss functionMSE on power outputMSE on rendered vs real silhouette pixels
OptimizerAdam, lr 1×10⁻⁴Adam (also used for control, lr 0.04)
Data headacheMissing/faulty sensor readingsClass imbalance: mostly-black images
Honest evaluationBootstrap confidence intervals, weather-type breakdownRS/NN baselines, 2,000-sample test set
Adaptation storyClustering to avoid retraining 84 modelsFine-tuning to recover from damage

Master the shared machinery once (Modules 1–8), then each paper becomes a case study (Modules 9–10) rather than a mountain.

Exam warm-up — say it out loud Before taking the quiz, answer these aloud in under a minute each, with no notes:
  1. Why does intermittent solar power threaten grid stability, and how does forecasting help?
  2. What is the difference between a macro-level and an aggregated inverter-level forecast?
  3. What can a robot with a learned self-model do that a robot with fixed CAD kinematics cannot?
  4. In one sentence each: what is Paper 1's finding? What is Paper 2's contribution?

Module 0 Quiz

10 questions. Aim for 80%+ before moving to Module 1.