| name | spiking-arm-locomotor-coordination |
| description | Spiking Neural Network architecture coordinating bipedal locomotion and arm control via NEF/SPA with biologically grounded basal ganglia for humanoid robots |
| tags | ["spiking neural network","neuromorphic","humanoid control","basal ganglia","locomotion","arm control","NEF","SPA"] |
| version | 1 |
| arxiv_id | 2606.11034v1 |
| created | 2026-06-10T00:00:00.000Z |
A Spiking Neural Architecture for Coordinating Arm and Locomotor Control
Paper Information
- arXiv ID: 2606.11034v1
- Authors: Lea Steffen, Kathryn Simone, Graeme Damberger, Travis DeWolf, Hudson Ly, Chris Eliasmith
- Published: 2026-06-09
- Categories: cs.RO, cs.NE
- URL: https://arxiv.org/abs/2606.11034v1
Summary
This paper presents the first integrated spiking controller combining bipedal locomotion and arm control on a full-scale humanoid platform. Using the Neural Engineering Framework (NEF) and Semantic Pointer Architecture (SPA), the system coordinates force-based arm control and locomotion mediated by a biologically grounded spiking basal ganglia model, enabling future deployment on low-power neuromorphic hardware.
Key Contributions
1. Integrated Spiking Architecture
- First full-scale humanoid: Combines both locomotor and arm control in single spiking system
- Action selection: Basal ganglia mediates switching between walking and arm control
- Force-based arm control: Novel approach using force feedback for manipulation
- Bipedal locomotion: Path-following walking with stability
2. Biological Grounding
- Basal ganglia model: Spiking implementation of disinhibition-based action selection
- Cortical circuits: Motor cortex regions for arm and locomotion control
- Sensorimotor integration: Tactile and proprioceptive feedback in spiking circuits
3. Neuromorphic Implementation
- Nengo framework: Neural Engineering Framework for spiking network construction
- Isaac Sim validation: Co-simulation of neural control with physics engine
- Energy efficiency: Designed for low-power neuromorphic hardware deployment
Methodology Details
Neural Engineering Framework (NEF)
- Representation: Neural populations encode vectors via distributed firing rates
- Transformation: Computations via weighted connections between populations
- Dynamics: Temporal processing via recurrent connections and synaptic filters
Semantic Pointer Architecture (SPA)
- High-level control: Symbolic action representations in neural substrate
- Bind/unbind operations: Associative memory for action sequencing
- Routing: Basal ganglia disinhibition controls information flow
Basal Ganglia Action Selection
Cortex → Striatum (Go/NoGo pathways)
↓
GPi/SNr (Output nucleus)
↓
Thalamus (Disinhibition)
↓
Motor cortex (Selected action execution)
System Components
1. Locomotion Controller
- Bipedal walking: Dynamic balance and path-following
- Foot placement: Adaptive stepping based on terrain
- Gait generation: Central pattern generator + sensory modulation
2. Arm Controller
- Force-based control: Target reaching via force feedback
- Digit drawing: Continuous trajectory generation
- Manipulation: Object interaction with tactile sensing
3. Action Selection System
- Basal ganglia: Spiking model with striatum, GPi/SNr, thalamus
- Cortical routing: Motor cortex regions for selected action
- Switching mechanism: Disinhibition enables rapid action transitions
Experimental Validation
Demonstrated Tasks
- Target reaching: Arm reaches to specified locations
- Digit drawing: Continuous drawing of digits (0-9)
- Path-following locomotion: Walking along predefined paths
- Action switching: Transitioning between walking and arm control
Co-Simulation Platform
- Nengo: Neural network simulation (spiking dynamics)
- Isaac Sim: Physics simulation (humanoid robot dynamics)
- Integration: Real-time communication between neural controller and physics engine
Technical Details
Spiking Implementation
- Neuron model: LIF (Leaky Integrate-and-Fire) neurons
- Encoding: Rate coding with tuning curves
- Synaptic filters: Exponential post-synaptic currents
- Time steps: 1ms simulation resolution
Force-Based Arm Control
- Force feedback: Simulated tactile sensors on fingertips
- Target specification: Desired force profile for reaching
- Error correction: Online adaptation via sensory feedback
- Joint control: Force-to-joint-torque transformation
SPA Operations
- Bind: Circular convolution for associative memory
- Unbind: Inverse operation for memory retrieval
- Cleanup: Similarity-based memory selection
- Routing: Disinhibition-based pathway selection
Applications
Primary Use Cases
- Humanoid robots: Full-body control with coordinated locomotion and manipulation
- Neuromorphic platforms: Energy-efficient deployment on specialized hardware
- Prosthetics: Bio-inspired control for assistive devices
- Assistive robotics: Adaptive systems for human-robot collaboration
Advantages vs. Traditional Approaches
- Energy efficiency: Spiking networks consume less power than ANN
- Biological plausibility: Grounded in neuroscience principles
- Integrated control: Single unified framework for multiple actions
- Online adaptation: Learning through sensory feedback
Future Directions
Hardware Deployment
- Intel Loihi: Neuromorphic chip implementation
- SpiNNaker: Massively parallel spiking simulation
- Braindrops: Analog neuromorphic processors
Extensions
- Multi-modal sensing: Vision, auditory, proprioceptive integration
- Learning: On-chip plasticity for skill acquisition
- Social interaction: Human-robot collaborative tasks
- Terrain adaptation: Outdoor walking on uneven surfaces
Implementation Notes
Key Innovations
- First integrated control: Previous SNN systems addressed locomotion or arm separately
- Basal ganglia mediation: Biologically grounded action selection mechanism
- Force-based approach: Novel arm control paradigm using force feedback
Challenges Addressed
- Action coordination: Switching between locomotor and manipulator modes
- Real-time control: Low-latency spiking processing for dynamic tasks
- Stability: Balance during locomotion and manipulation
Related Skills
snn-learning-survey: Overview of SNN training methodsneuromorphic-supremacy-hybrid-astrocytic-spiking: Advanced neuromorphic architecturesrobotic-locomotion-dynamics: Locomotion biomechanicsbci-motor-decoding: Motor cortex decoding for prosthetics
References
Activation: Use when designing spiking controllers for humanoid robots, implementing integrated locomotion-manipulation systems, building biologically grounded basal ganglia models, or deploying neural control on neuromorphic hardware. Keywords: spiking humanoid, basal ganglia, locomotion arm coordination, NEF SPA, force-based control, Nengo Isaac Sim, neuromorphic robotics.
