Project Echo
Robot learning
Building the first humanoid in Delft. The first step towards this goal is developing the first half-scale prototype.
Using computer vision, the robot detects the ball through colour detection and calculates its trajectory. Object detection and ArUco tags are used to perceive and map the surrounding environment. The arm movements are controlled and trained through a combination of imitation learning and reinforcement learning.
Status
Current Prototype - ECHO I:
The active system consists of two modular robotic arms mounted on a support frame, designed to catch and throw a volleyball. The robot uses computer vision, ArUco tags, and reinforcement learning to map its environment and control its movements.
Industry Pilot Projects:
Collaborating with Ambionic (a Skelex spin-off) to test a new single-rintable compliant robotic hand on the ECHO arm.
Building a cost-efficient, open-source hand for Daimon Robotics to demonstrate their DM-Tac F vision-based tactile sensors.
Recent Milestones:
Project Echo won the Student Award at the Regional Robotics Awards 2026 and is advancing to the national final at FuturePulse 2026 in Eindhoven.
Context
The Problem:
Current reliance on foreign, closed-architecture humanoids (such as those from Unitree) restricts hardware modification and complicates repair for university research groups.
The Solution:
Project Echo aims to provide a local, open-source humanoid platform with nearby support for researchers and student teams in Delft
The Approach:
Rather than treating a humanoid as a fixed unit, it is designed as a configurable, modular platform where researchers can adapt specific subsystems (arms, legs, hands) to their exact research needs.
Roadmap
Development Strategy:
The project is splitting into separate upper and lower body developments to iterate on manageable steps before full integration.
Upper Body (Arms & Torso):
The next iteration will be approximately 1.60m tall, featuring an upgraded mechanical design. The team is replacing hobby servos with better-suited actuators and replacing hand-soldered electronics with custom PCBs.
Hands:
The new hand design will feature five independently actuated fingers on a 2-DoF wrist. It will transition from servo motors to linear actuators and integrate Daimon Robotics sensors directly into the fingertips.
Lower Body (Legs):
The legs team is building the first prototype using CubeMars actuators. The initial focus is on achieving stable walking and performing dynamic motions like squats, which will then be benchmarked against other open-source projects like LeRobo.