Path planning on the TrueNorth neurosynaptic system

Fischl, Kate D.; Fair, Kaitlin; Tsai, Wen-Fa; Sampson, Jack; Andreou, Andreas G.

doi:10.1109/iscas.2017.8050932

Cited by 11 publications

(4 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While most demonstrations of the SNN advantages focus on the energy gains and come at a cost of a drop in performance [9], [11], [12], this is perhaps the first time that an energy-efficient method also demonstrates better accuracy than the current state of the art, at least in the tested robotic navigation tasks. The accuracy increase is partly due to our hybrid training approach that helped overcome the limitation of SNNs in representing high precision values, which led to better optimization.…”

Section: Discussionmentioning

confidence: 89%

“…brain-inspired alternative architecture to deep neural networks (DNN) in which neurons compute asynchronously and communicate through discrete events called spikes [8]. We and others have recently shown how the realization of SNNs in a neuromorphic processor results in low-power solutions for mobile robots, ranging from localization and mapping of mobile robots [9] on Intel's Loihi [10] to planning [11] and control [12]. For mapless navigation, most SNN-based approaches employ a reward modulated learning, where a global reward signal drives the local synaptic weight updates [13], [14].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Reinforcement co-Learning of Deep and Spiking Neural Networks for Energy-Efficient Mapless Navigation with Neuromorphic Hardware

Tang

Kumar

Michmizos

2020

Preprint

View full text Add to dashboard Cite

Energy-efficient mapless navigation is crucial for mobile robots as they explore unknown environments with limited on-board resources. Although the recent deep reinforcement learning (DRL) approaches have been successfully applied to navigation, their high energy consumption limits their use in many robotic applications. Here, we propose a neuromorphic approach that combines the energy-efficiency of spiking neural networks with the optimality of DRL to learn control policies for mapless navigation. Our hybrid framework, Spiking deep deterministic policy gradient (SDDPG), consists of a spiking actor network (SAN) and a deep critic network, where the two networks were trained jointly using gradient descent. The trained SAN was deployed on Intel's Loihi neuromorphic processor. The co-learning enabled synergistic information exchange between the two networks, allowing them to overcome each other's limitations through a shared representation learning. When validated on both simulated and real-world complex environments, our method on Loihi not only consumed 75 times less energy per inference as compared to DDPG on Jetson TX2, but also had a higher rate of successfully navigating to the goal which ranged by 1% to 4.2%, depending on the forward-propagation timestep size. These results reinforce our ongoing effort to design braininspired algorithms for controlling autonomous robots with neuromorphic hardware.

show abstract

Section: Discussionmentioning

confidence: 89%

Section: Introductionmentioning

confidence: 99%

Reinforcement co-Learning of Deep and Spiking Neural Networks for Energy-Efficient Mapless Navigation with Neuromorphic Hardware

Tang

Kumar

Michmizos

2020

Preprint

View full text Add to dashboard Cite

show abstract

“…Unlike other path planners, the spiking wavefront algorithm is a network of spiking neurons. Therefore, the algorithm is compatible with power efficient neuromorphic hardware, as was shown by implementing it on the IBM TrueNorth NS1e [Fischl et al, 2017]. The spiking wavefront propagation algorithm is also supported by the observation of spreading activation of hippocampal place activity prior to taking action [Dragoi and Tonegawa, 2011, Pfeiffer and Foster, 2013].…”

Section: Discussionmentioning

confidence: 99%

Flexible Path Planning Through Vicarious Trial and Error

Krichmar

Ketz

Pilly

et al. 2021

Preprint

View full text Add to dashboard Cite

Flexible planning is necessary for reaching goals and adapting when conditions change. We introduce a biologically plausible path planning model that learns its environment, rapidly adapts to change, and plans efficient routes to goals. Unlike prior models of hippocamapl replay, our model addresses the decision-making process when faced with uncertainty. We tested the model in simulations of human and rodent navigation in mazes. Like the human and rat, the model was able to generate novel shortcuts, and take detours when familiar routes were blocked. Similar to rodent hippocampus recordings, the neural activity of the model resembles neural correlates of Vicarious Trial and Error (VTE) during early learning or during uncertain conditions. Similar to rodent studies, after learning, the neural activity resembles forward replay or preplay predicting a future route, and VTE activity decreases. We suggest that VTE, in addition to weighing possible outcomes, is a way in which an organism may gather information for future use.

show abstract

“…The parallel nature of the algorithm makes it particularly suitable for neuromorphic processors which are inherently parallel. The wavefront path planner has previously been implemented successfully on neuromorphic hardware, such as IBM's TrueNorth processor [27].…”

Section: Motor Modulementioning

confidence: 99%

A neuromorphic electronic artist for robotic painting

Schürmann,

D'Angelo,

Indiveri

et al. 2024

Preprint

View full text Add to dashboard Cite

Recent advances in computer vision and deep learning have led to a surge of interest in the field of AI-generated art, including digital image creation and robot-assisted painting. Traditional painting machines rely on static images and offline processing to incorporate visual feedback into their painting process. However, this approach does not consider the dynamic nature of painting and fails to decompose complex overlapping patterns into individual strokes. As an alternative to frame-based RGB cameras, neuromorphic cameras capture changes in light intensity within a scene via asynchronous event streams, promising to overcome some of the inherent limitations of traditional computer vision techniques. In this project, a robotic system for physical painting is presented which utilizes event-based visual input from a Dynamic Vision Sensor (DVS) camera. To take advantage of the camera's ultra-low latency and sparse encoding, the proposed system also employs event-based information processing, implemented with spiking neural networks on the neuromorphic DynapSE-1 processor. The robotic system receives DVS sensory data which represents the trajectory of a brush stroke and computes the required joint velocities to recreate the stroke with a 6-DOF robotic arm in a closed-loop manner. The controller additionally integrates tactile feedback from a force-torque sensor to dynamically adjust the end-effector’s distance towards the canvas depending on the brush’s deformation. Within the scope of the project, it was further demonstrated how speed information about a perceived brush stroke can be extracted from DVS data. The system was tested in a real-world setting and successfully generated a collection of physical brush strokes. The proposed network is a first step towards a fully spiking robotic controller with the ability to seamlessly incorporate event-based sensory feedback, providing ultra-low latency responsiveness. Beyond its utility in robot-assisted painting, the developed network is applicable to any robotic task requiring real-time adaptive control.

show abstract

Path planning on the TrueNorth neurosynaptic system

Cited by 11 publications

References 17 publications

Reinforcement co-Learning of Deep and Spiking Neural Networks for Energy-Efficient Mapless Navigation with Neuromorphic Hardware

Reinforcement co-Learning of Deep and Spiking Neural Networks for Energy-Efficient Mapless Navigation with Neuromorphic Hardware

Flexible Path Planning Through Vicarious Trial and Error

A neuromorphic electronic artist for robotic painting

Contact Info

Product

Resources

About