Pavan Velagaleti
← Home

Articulus Surgical Robotics

Cosmos architecture powering Pulsar: real-time control, teleoperation, and safety-aware motion

2023–2025 Tier 1 Lead Robotics Software Engineer — Architect of Cosmos (Reporting to CEO)
Articulus Surgical Robotics
Architecture note: Pulsar is the surgical robotic platform. Cosmos is the real-time control and communication architecture I designed and implemented to run it.
Confidentiality: This page shows representative architecture only. Details are generalized to respect confidentiality.

What I built

Cosmos is a real-time control and communication architecture for multi-DOF teleoperated surgical robots such as Pulsar. It supports deterministic tool-frame motion (RPY + insertion), dynamic instrument swaps, safety-aware state gating, and low-latency teleoperation under strict timing and reliability constraints. The architecture spans high-level teleoperation and kinematics down to FPGA-backed BLDC motor control, integrating commercial motor-control platforms such as ODrive and mjbots moteus for deterministic actuation, closed-loop stability, and fault-safe behavior. The software was designed with medical device software lifecycle expectations (ISO 62304) and relevant IEC electronics standards in mind.

Problem

Architect deterministic real-time control and communication pipelines for a multi-DOF surgical robotics platform, supporting tool-frame orientation and insertion control (RPY + linear tool-axis motion), dynamic instrument swaps, teleoperated micro/macro motion, and safety-aware behavior aligned with medical device software and electronics standards.

Approach

  • Architected Cosmos’ core real-time pipeline: feedback ingestion → synchronized control loops → deterministic actuation
  • Designed multithreaded execution models with explicit scheduling and timing guarantees
  • Integrated DDS middleware (Cyclone DDS, Fast DDS) for low-latency, reliable communication across robotic subsystems
  • Designed software structures aligned with ISO 62304 medical device lifecycle expectations
  • Implemented safety and constraint handling: limits, saturation, fault responses, standards-aware workflows
  • Built tool-frame motion control (RPY + insertion) and integrated FK/IK into teleoperation pipelines
  • Designed instrument lifecycle handling: identification, frame updates, safety re-initialization
  • Built observability tooling: high-rate logging and dashboards for bring-up, debugging, and validation
  • Designed and tuned BLDC motor control loops (current, velocity, position) using FOC for surgical manipulators
  • Integrated and validated commercial BLDC motor controllers (e.g., ODrive, mjbots moteus) within a real-time surgical control architecture.
  • Developed FPGA-based real-time motor-control and IO pipelines (PWM generation, encoder decoding, safety interlocks)
  • Integrated FPGA and motor-controller actuation paths with Cosmos’ real-time C++ control stack, debugging jitter, race conditions, and timing faults across hardware–software boundaries

Engineering decisions

Deterministic execution
Explicit real-time scheduling boundaries to reduce jitter and simplify debugging across control and actuation layers.
Motor controller selection and abstraction
Evaluated and integrated ODrive and mjbots moteus to balance performance, observability, and safety, while abstracting actuation behind Cosmos’ real-time control interfaces.
Safety gating and recovery
State-driven enable/disable logic with predictable fault recovery paths spanning software, FPGA IO, and motor controllers.

Ownership

  • Architecture ownership of Cosmos scheduling, determinism, and loop timing
  • Direct collaboration with CEO on software strategy and delivery priorities
  • Technical leadership across control architecture, computer vision, electronics, and hardware teams
  • Designed software workflows aligned with ISO 62304 expectations
  • Teleoperation behavior and tool-frame motion pipelines
  • Ownership of BLDC motor control behavior, tuning, and closed-loop performance
  • Integration and validation of ODrive and mjbots moteus motor controllers
  • FPGA-based real-time IO, motor actuation pipelines, and safety interlocks
  • Safety constraints, fault responses, and lifecycle re-initialization
  • Observability tooling for validation and live demos

Results

  • ~1–3 ms control-cycle compute budget under RT scheduling
  • ~22–26 ms end-to-end command → motion latency
  • Stable closed-loop BLDC actuation using ODrive and mjbots moteus under deterministic real-time constraints
  • Safe dynamic instrument swaps without loss of control continuity
  • Faster bring-up and debugging through integrated observability tooling

Gallery

Stack

C++Real-timeControlMotor ControlBLDCFOCFPGAODrivemjbots moteusDDSSafetySystems