Modular Legged Robots
Modbot: Reconfigurable Legged Robot Modules
An 8-DoF biped module that walks alone or links into a quadruped — built to measure what modularity really costs in reliability, control, and energy.
- Period
- 2026 – present ● active
- Platform
- Modbot (custom biped module) · MuJoCo
- Methods
- Context-Conditioned RL · Domain Randomization · Cost-of-Transport Benchmarking
- With
- Panos Trahanias (FORTH)
Overview
Modbot is my current research platform at FORTH: a compact 8-DoF biped locomotion module that is self-sufficient on its own and can link with a partner module into a quadruped. The project’s goal is not another universal modular platform — it is the measurement the field has skipped: treating modularity as generality purchased at a cost, and quantifying that cost.
Research problem
Every degree of modular freedom is taxed three times: mechanically (each interface is a new failure point), computationally (each configuration axis is another dimension the controller must cover), and energetically (unused optionality is permanent mass and watts). These three currencies are almost never put on one ledger, and the decisive benchmark — a linked-biped quadruped raced head-to-head against a monolithic equivalent — does not exist in the published literature. Modbot is built to run that race.
System
- Hardware model: an 8-DoF biped with hip, thigh, and calf links per leg, designed in CAD and exported to URDF/MJCF for large-scale simulation.
- Control: configuration-conditioned reinforcement-learning policies trained in GPU simulation across the module family — single modules, foot and payload variants, and linked-quadruped configurations.
- Evaluation: a common ledger of velocity tracking, disturbance rejection, and cost of transport across the configuration ladder — the generality–cost frontier.
Context
The project sits inside the RAICAM doctoral network (Robotics and AI for Critical Asset Monitoring), where mobile robots inspect and maintain critical infrastructure — a mission profile in which one adaptable platform competing against several purpose-built ones is a real procurement question, not a thought experiment.
Status
Active. The module family is modeled and training in simulation; a manuscript on the generality–cost frontier and the biped-linkage benchmark is in preparation (2026).