Echo Bot

Echo Bot gives the Echo architecture a physical body. A cat-sized hexapod, six legs, eighteen motors, sensors, vision, voice, and an onboard brain. Here, decisions deal with mass, balance, effort, energy, and wear. The robot perceives, maintains its internal state, and adapts its movements to the real world instead of replaying a choreography.

The Echo robot in 3D, a controllable hexapod: you can rotate it, look at it from every angle, and talk to it. Rotate Echo in 3D, throw the ball to him, and watch how he reacts to the butterflies.

Technical sheet

General information

Architect / Creator: David Pichot

Development: Claude Code

Started: June 2026

Development time: 53 days

First physical project built end to end

Mechanics

Morphology: hexapod, 6 legs × 3 joints

Motors: 18 × ST3215, 30 kg·cm, load and temperature feedback

Standing mass: 7 kg

Design: Cinema 4D, PLA-printed parts

Retractable tray for the power board

Brain and perception

Compute: Raspberry Pi 5 16 GB + AI HAT+ 2 (Hailo-10H, 40 TOPS)

Memory: Crucial X10 SSD, 1 TB

Vision: OV5647 IR-CUT camera and 2 infrared projectors

Screen: 7-inch AMOLED panel

Voice: 2 INMP441 microphones, 2 amplified speakers

Sensors

Distance: ToF VL53L1X

Presence: LD2410 radar

Balance: MPU6050 inertial measurement unit

Load: each motor reports its load and temperature

Energy: INA226 current monitor on a 50 A shunt

Energy

Battery: 3S 5200 mAh LiPo, 11.1 V motor rail

Protection: 3S 40 A BMS, 30 A resettable circuit breaker, anti-spark cutoff

Distribution: two separate 5 V power supplies, brain and peripherals

Cooling: 5 fans driven by temperature

Structural innovations

Homeostasis: drifting needs, not a score to maximize

Three decision levels: reflex, learned movement, reflection

Walking by leg role, balance never broken

Body pulled by the legs, never slid across the floor

Hybrid landing absorption: motors and springs in parallel

Transfer from simulation to the physical robot has yet to be validated

Status

Project status: In progress

Problem

An intelligence that lives only on a screen operates in a world written entirely for it. It does not fall, does not get tired, does not overheat, does not break anything, and its errors can always be absorbed by software. Even when it learns, it learns in an environment that offers no independent reality in return. I wanted to know what remains of my architectures when they are given a body and have to deal with something they do not control: mass, eighteen imperfect motors, a battery that drains, friction, inertia, heat, obstacles, and physical consequences.

From there, the problem is no longer just to make a robot walk or execute commands. I have to build continuity between perception, internal state, decision, and movement. Fast enough to produce a reflex when needed. Open enough to learn when the answer does not exist yet. Grounded enough in the real world for failure to have a cost. Echo Bot came from this question: what does an intelligence become when it can no longer cheat with the world?

Answer

Echo Bot combines an assembled and wired physical hexapod with a control architecture studied in simulation. Sensors and motor feedback are intended to inform an internal state: load, temperature, balance and energy can then contribute to decisions. Autonomous walking on the physical robot remains to be validated.

Decisions are organized by urgency: local reflexes, learned movements and deliberation for new situations. Simulation is used to explore these behaviors before testing them on the robot. The aim is to build a system that accounts for the real constraints of its body.

Features

Echo Bot is not a set of functions placed side by side. The mechanics, sensors, electronics, internal state, and decision layers were designed as one system: every problem the body meets creates new information, every piece of information can change a decision, and every decision eventually returns to face the real world. The seven blocks below show where this architecture becomes concrete.

One articulated leg of the Echo robot: you can grab the foot or the base and manipulate it. Drag the foot or the base.

Adaptive hexapod gait

Echo’s gait is not a sequence of recorded poses. Each leg is assigned a role that changes during movement. Four points of contact stay on the ground as much as possible, and the body moves forward because the legs pull and push it, not because it is artificially translated above the floor. A leg that is already well placed has no reason to lift: it keeps working until moving it becomes necessary. This logic reduces wasted effort, maintains balance, and, above all, makes motion depend on the body’s real state rather than on a predefined choreography.

The visual isolates one leg to show this mechanism without the rest of the robot. Grab the foot or the base and move them: the joints follow their real constraints, and the areas taking the load appear in cyan. You can see directly how foot position, leg extension, and body movement change the mechanical constraints.

The Echo robot’s seven internal needs, live: they drift, the one that falls out of range takes over, and you can read the resulting action and its cause. Change one of the seven internal gauges and watch its effect on the other needs and the chosen action.

Robot internal needs

Echo does not choose its actions from a list of predefined behaviors, but from seven internal needs, energy, integrity, social contact, rest, fear, curiosity, and frustration, that drift continuously. Each has a balance zone, a weight, and a rate of change. As long as the system stays close to that zone, nothing really dictates its behavior; as soon as one need moves far enough away, its importance rises until it can take priority over the others. Mood and behavior therefore emerge from the system’s overall state, rather than from a direct order like “if X, then Y”.

The visual runs exactly the same model as the robot. You can trigger a presence, a stroke, or an impact, or move one of the gauges directly and immediately see how the balance shifts, which need becomes dominant, and what behavior follows from it. One hour of life is compressed into a few seconds to make visible changes that, on Echo, normally happen much more slowly.

Diagram of Echo’s cognitive architecture: sensors, percepts, reflex loop, seven-gauge homeostasis, two-gate language, outputs. Trace a signal from the sensors through reflexes, learned actions, or reasoning to the robot’s outputs.

Reflexes, gestures, reflection

Not all of Echo’s decisions have the same urgency or the same cost. The architecture is organized into three levels: reflexes immediately handle anything related to the body and safety, learned gestures take over situations that are already known, and reflection only comes in when a problem remains ambiguous or new. The goal is to never mobilize a costly layer for something a simpler mechanism already knows how to do. Through learning, a response that first required reflection can gradually become a gesture, then an automatic action.

The diagram lets you explore the full architecture, from the sensors to the system outputs. Zoom in, move through the brain, and follow the connections: you can see the three levels, homeostasis, the different families of perception, and the gates that decide how far information must travel upward before a response is produced.

The robot’s head open: the Raspberry Pi 5, the ribbon of wires, and the control board housed behind the hatch

Robot’s onboard brain

Echo carries most of its computation in its head. The Raspberry Pi 5 orchestrates the system, the dedicated accelerator handles vision processing, and the SSD stores local memory. The camera, the screen, the audio interfaces, and the different links are grouped in the same volume so the robot stays autonomous and only a few connections pass through the neck to the rest of the body. This integration constraint forced me to think of the electronics as a physical architecture, not as a list of boards to stack.

The visual shows the head open, with the main components in place. Look at how the Raspberry Pi, the HAT, the camera, and the screen are arranged in a very small space: the point is not only to make everything fit, but to preserve access to the boards, airflow, cable paths, and the ability to take the assembly apart without rebuilding the robot.

The complete electrical diagram of the robot: battery, protection, regulation, two control boards, eighteen servomotors on six legs, and the path of every wire. Trace power and signals between the battery, control boards, and the robot’s eighteen motors.

Robot wiring diagram

Echo’s electronics were designed before they were wired. Every power supply, sensor, motor, board, and connection was first placed in a single schematic that serves as the reference for the physical build. This method became essential as the system grew more complex. It lets me check voltages, grounds, signal paths, and dependencies before a mistake can damage a component. It was also the most unforgiving part of the project, because I started with no practical knowledge of electronics.

The visual shows the wiring schematic as it is actually used in the project. Zoom in and move through the circuit to follow the links between the different subsystems: power, compute, sensors, motors, and peripherals. This is not an illustration rebuilt for the page. It is the design document used to understand and verify the robot before plugging it in.

The Echo robot in sketch mode: push it, press on it, and each joint changes color according to what its motor is taking on. Push or press the robot: its joints change colour according to the load they bear.

Simulation of motor loads

Echo must do more than know how to move; it must know what its movements cost its own body. Each motor reports its load, the contact points absorb part of the weight, and some areas gradually become strained, then dangerous. This information prevents the system from treating a pose as correct just because it is geometrically possible. When a constraint becomes too high, the robot must be able to yield, move a contact point, or abandon a position instead of pushing on until something breaks.

The visual shows this link between posture and load directly. Push the body or one of the legs and watch the reaction: the joints change color according to the load demanded from the motors, while the feet and damping systems show what they are absorbing. The displayed values show when the robot stays within a comfortable range, when it begins to strain, and when it prefers to lose a position rather than exceed its mechanical limits.

The robot standing on the desk, in front of the design screen
The assembled robot, wiring exposed, before the shell is closed
The printed legs lined up on the desk before assembly
The pencil sketch of the leg, joint by joint
The 3D printer producing one of the robot’s parts
The production cockpit: the print queue with the current part and its remaining time, the camera pointed at the printer, the tree of parts to model, and the print and assembly schedule
The damping spring seated in its printed sleeve
The robot’s platform, drilled, lightened, and screwed in place
The six finished legs, placed side by side
The body floor with its fans and wiring
The assembled body, shell and legs in place
A close view of the mounted joints

Building the robot

Echo is also my first robot built physically from end to end. I started with a 3D printer, very little equipment, and no real electronics experience. The mechanics went surprisingly well: almost every part worked from the first prints, except for the top of the coxa, which I had undersized and rebuilt as several pieces with crossed print orientations to handle the forces better. The real challenge came when I had to fit all the electronics into the body and head, then wire, crimp, solder, diagnose, and make the whole assembly serviceable. Echo gradually became as much a robot as a benchmark for building my workshop: soldering station, power supply, cables, connectors, screws, test components, servos, and soon CNC. Everything learned or bought for it must be useful for the machines that come next.

The slideshow follows that transformation, from the first leg sketch to the assembled and wired robot. Scroll through the steps or open a photograph to inspect it in detail: prints, springs, platform, legs, cooling, electronics, and final assembly show how a 3D model gradually became a seven-kilogram object able to house its own brain.

Next steps

The first step is much less spectacular than a new feature: finish Echo, close the body, make the electronics reliable, and make it walk for real. Until the robot has lived long enough in the physical world, part of the architecture remains a hypothesis. Walking, balance, force feedback, perception, energy, and body reactions now have to be tested together against what simulation never reproduces perfectly: mechanical play, friction, variations between motors, measurement errors, and the simple unpredictability of the real world.

Then the part that interests me most will probably begin: letting it learn to live with its own body. Its brain will have to gradually discover which actions work, which ones cost too much, how to move without wasting energy, how to react when a gesture fails, and how to turn a solution found through reasoning into faster behavior. The simulation will remain a test ground, but the real robot will continually correct it. What works in one will have to survive in the other.

Echo is not meant to be an isolated prototype. It is my first complete body, the one I can measure, break, correct, and learn from. Motor models, movement strategies, perception mechanisms, reflexes, and validated behaviors can then become reusable building blocks for my next robots. The long-term goal is not only for Echo to become more intelligent over time, but for each robot built after it to begin its life with part of what the previous ones have already learned.

Let's work together

Innovation grows out of collaboration. Together, let's build solutions that matter.
Owan
Hello, my name is Owan, I am David's assistant. How can I help you?
0/2000